Release 1.3.2

Upgrade Rust compiler in container image (1.81.0 -> 1.82.0)
Upgrade component services
2024-11-12 10:37:49 +02:00 · 2024-11-12 10:37:16 +02:00 · 2024-11-12 10:36:54 +02:00 · 2024-11-12 10:22:26 +02:00 · 2024-10-03 16:30:58 +03:00 · 2024-10-03 16:28:49 +03:00
62 changed files with 1368 additions and 648 deletions
--- a/CHANGELOG.md
+++ b/CHANGELOG.md
@@ -1,3 +1,35 @@
+# (2024-11-12) Version 1.3.2
+
+Dependency updates.
+
+
+# (2024-10-03) Version 1.3.1
+
+- (**Improvement**) Improves fallback user mentions support for old clients (like Element iOS) which use the bot's display name (not its full Matrix User ID). ([d9a045a5e4](https://github.com/etkecc/baibot/commit/d9a045a5e41d2b99694f92ec9e90f47529546d89))
+
+
+# (2024-10-03) Version 1.3.0
+
+**TLDR**: you can now use OpenAI's [o1](https://platform.openai.com/docs/models/o1) models, benefit from [prompt caching](https://platform.openai.com/docs/guides/prompt-caching) and mention the bot again from old clients lacking proper [user mentions support](https://spec.matrix.org/latest/client-server-api/#user-and-room-mentions) (like Element iOS).
+
+- (**Feature**) Introduces a new `baibot_conversation_start_time_utc` [prompt variable](./docs/configuration/text-generation.md#️-prompt-override) which is not a moving target (like the `baibot_now_utc` variable) and allows [prompt caching](https://platform.openai.com/docs/guides/prompt-caching) to work. All default/sample configs have been adjusted to make use of this new variable, but users need to adjust your existing dynamically-created agents to start using it. ([85e66406dc](https://github.com/etkecc/baibot/commit/85e66406dc6f430741c7819f420e2df4ae6e8d3b))
+
+- (**Improvement**) Allows for the `max_response_tokens` configuration value for the [OpenAI provider](./docs/providers.md#openai) to be set to `null` to allow [o1](https://platform.openai.com/docs/models/o1) models (which do not support `max_response_tokens`) to be used. See the new o1 sample config [here](./docs/sample-provider-configs/openai-o1.yml). ([db9422740c](https://github.com/etkecc/baibot/commit/db9422740ceca32956d9628b6326b8be206344e2))
+
+- (**Improvement**) Switches the sample configs for the [OpenAI provider](./docs/providers.md#openai) to point to the `gpt-4o` model, which since 2024-10-02 is the same as the `gpt-4o-2024-08-06` model. We previously explicitly pointed the bot to the `gpt-4o-2024-08-06` model, because it was much better (longer context window). Now that `gpt-4o` points to the same powerful model, we don't need to pin its version anymore. Existing users may wish to adjust their configuration to match. ([90fbad5b64](https://github.com/etkecc/baibot/commit/90fbad5b643cd06c23179f055a309ec6a7cba161))
+
+- (**Bugfix**) Restores fallback user mentions support (via regular text, not via the [user mentions spec](https://spec.matrix.org/latest/client-server-api/#user-and-room-mentions)) to allow certain old clients (like Element iOS) to be able to mention the bot again. Support for this was intentionally removed recently (in [v1.2.0](#2024-10-01-version-120)), but it turned out to be too early to do this. ([b40226826f](https://github.com/etkecc/baibot/commit/b40226826fe914d0d5d265230ebc5bac8058b6f7))
+
+
+# (2024-10-01) Version 1.2.0
+
+- (**Feature**) Adds support for [on-demand involvement](./docs/features.md#on-demand-involvement) of the bot (via mention) in arbitrary threads and reply chains ([9908512968](https://github.com/etkecc/baibot/commit/990851296828168c2106eb3f4668833e9e5a7463)) - fixes [issue #15](https://github.com/etkecc/baibot/issues/15)
+
+- (**Improvement**) Simplifies [Transcribe-only mode](./docs/features.md#transcribe-only-mode) reply format (removing `> 🦻` prefixing) to allow easier forwarding, etc. ([e6aa956423](https://github.com/etkecc/baibot/commit/e6aa95642376ee7d87932d0e66dcfedf261b188b)) - fixes [issue #14](https://github.com/etkecc/baibot/issues/14)
+
+- (**Bugfix**) Fixes speech-to-text replies rendering incorrectly in certain clients, due to them confusing our old reply format with [fallback for rich replies](https://spec.matrix.org/v1.11/client-server-api/#fallbacks-for-rich-replies) ([e6aa956423](https://github.com/etkecc/baibot/commit/e6aa95642376ee7d87932d0e66dcfedf261b188b)) - fixes [issue #17](https://github.com/etkecc/baibot/issues/17)
+
+
 # (2024-09-22) Version 1.1.1

 - (**Bugfix**) Fix thread messages being lost due to lack of pagination support ([d4ddd29660](https://github.com/etkecc/baibot/commit/d4ddd29660d9f51d248119dd6032e68ab29e7d35)) - fixes [issue #13](https://github.com/etkecc/baibot/issues/13)
--- a/Cargo.lock
+++ b/Cargo.lock
--- a/Cargo.toml
+++ b/Cargo.toml
@@ -7,7 +7,7 @@ license = "AGPL-3.0-or-later"
 readme = "README.md"
 keywords = ["matrix", "chat", "bot", "AI", "LLM"]
 include = ["/etc/assets/baibot-torso-768.png", "/src", "/README.md", "/CHANGELOG.md", "/LICENSE"]
-version = "1.1.1"
+version = "1.3.2"
 edition = "2021"

 [lib]
@@ -17,7 +17,7 @@ path = "src/lib.rs"
 [dependencies]
 anthropic-rs = "0.1.*"
 anyhow = "1.0.*"
-async-openai = "0.24.*"
+async-openai = "0.26.*"
 base64 = "0.22.*"
 chrono = { version = "0.4.*", default-features = false, features = ["std", "now"] }
 # We'd rather not depend on this, but we cannot use the ruma-events EventContent macro without it.
@@ -26,13 +26,13 @@ mxidwc = "1.0.*"
 mxlink = ">=1.3.0"
 etke_openai_api_rust = "0.1.*"
 quick_cache = "0.6.*"
-regex = "1.10.*"
+regex = "1.11.*"
 serde = { version = "1.0.*", features = ["derive"], default-features = false }
 serde_json = "1.0.*"
 serde_yaml = "0.9.*"
-tempfile = "3.12.*"
-tiktoken-rs = { version = "0.5.*", features = ["async-openai"] }
-tokio = { version = "1.40.*", features = ["rt", "rt-multi-thread", "macros"] }
+tempfile = "3.14.*"
+tiktoken-rs = { version = "0.6.*", features = ["async-openai"] }
+tokio = { version = "1.41.*", features = ["rt", "rt-multi-thread", "macros"] }
 tracing = "0.1.*"
 tracing-subscriber = { version = "0.3.*", features = ["env-filter"] }
 url = "2.5.*"
--- a/2
+++ b/2
@@ -4,7 +4,7 @@
 #                                     #
 #######################################

-FROM docker.io/rust:1.81.0-slim-bookworm AS build
+FROM docker.io/rust:1.82.0-slim-bookworm AS build

 RUN apt-get update && apt-get install -y build-essential pkg-config libssl-dev libsqlite3-dev

--- a/Dockerfile.ci
+++ b/Dockerfile.ci
@@ -4,7 +4,7 @@
 #                                     #
 #######################################

-FROM docker.io/rust:1.81.0-slim-bookworm AS build
+FROM docker.io/rust:1.82.0-slim-bookworm AS build

 RUN apt-get update && apt-get install -y build-essential pkg-config libssl-dev libsqlite3-dev

--- a/README.md
+++ b/README.md
@@ -41,7 +41,7 @@ It's influenced by [chaz](https://github.com/arcuru/chaz), but does **not** use

 ![Introduction and general usage](./docs/screenshots/introduction-and-general-usage.webp)

-You can find more screenshots on the the [🌟 Features](./docs/features.md) and other [📚 Documentation](./docs/README.md) pages, as well as in the [docs/screenshots](./docs/screenshots) directory.
+You can find more screenshots on the [🌟 Features](./docs/features.md) and other [📚 Documentation](./docs/README.md) pages, as well as in the [docs/screenshots](./docs/screenshots) directory.


 ## 🚀 Getting Started
--- a/docs/access.md
+++ b/docs/access.md
@@ -16,6 +16,7 @@ Users:

 - ✅ can **invite the bot to rooms**
 - ✅ can **use all the bot's [features](./features.md)** ([💬 Text Generation](./features.md#-text-generation), [🦻 Speech-to-Text](./features.md#-speech-to-text), etc.) by sending room messages
+- ✅ can **mention the bot** in threads and reply chains to provoke it to respond to non-user messages (see [🌟 Features / 💬 Text Generation / On-demand involvement](./features.md#on-demand-involvement))
 - ✅ can **change the bot's configuration in a room** (e.g. `!bai config room ...` commands)
 - ❌ cannot **change the bot's global configuration** (e.g. `!bai config global ...` commands)
 - ❌ cannot **create new [🤖 Agents](./agents.md)** (neither in rooms, nor globally). See [💼 Room-local agent managers](#-room-local-agent-managers) for controlling which users can create agents.
--- a/docs/configuration/text-generation.md
+++ b/docs/configuration/text-generation.md
@@ -13,7 +13,7 @@ You may also wish to see:

 In Direct Message rooms with the bot (1:1 rooms), it most usually makes sense for the bot to respond to **all** of your messages, as shown on this [🖼️ screenshot](../screenshots/text-generation.webp).

-In group rooms (with multiple users), it may be more appropriate for the bot to only respond to messages that are **prefixed** with the command prefix (e.g. `!bai`), so that other chat exchange in the room will not trigger it. Such a setup is shown on this [🖼️ screenshot](../screenshots/text-generation-prefix-requirement.webp).
+In group rooms (with multiple users), it may be more appropriate for the bot to only respond to messages that are **prefixed** with the command prefix (e.g. `!bai`) or which are [mentioning](https://spec.matrix.org/latest/client-server-api/#user-and-room-mentions) the bot (e.g. `@baibot`), so that other chat exchange in the room will not trigger it. Such a setup is shown on the [🖼️ On-demand involvement in the room](../screenshots/text-generation-prefix-requirement.webp) screenshot.

 There are exceptions to these rules, and you can configure the bot to respond only to prefixed messages in a 1:1 room, or to respond to all messages even in a multi-user group room.

@@ -27,7 +27,10 @@ By default, the bot is **auto-configured (upon joining a new room)** to use the

 Example: `!bai config room text-generation set-prefix-requirement-type command_prefix` (this can also be set globally, see [🛠️ Room Settings](./README.md#room-settings))

-Regardless of this configuration, **the bot will also respond to messages which directly [mention](https://spec.matrix.org/latest/client-server-api/#user-and-room-mentions) the bot** (e.g. `@baibot`), even if they are not prefixed. An example of this can be seen on this [🖼️ screenshot](../screenshots/text-generation-prefix-requirement.webp).
+Regardless of this configuration, **the bot will also respond to messages by allowed [👥 Users](../access.md#-users) which directly [mention](https://spec.matrix.org/latest/client-server-api/#user-and-room-mentions) the bot** (e.g. `@baibot`), even if they are not prefixed. An example of this can be seen on these screenshots:
+
+- [🖼️ On-demand involvement in a thread](../screenshots/text-generation-on-demand-thread-involvement.webp)
+- [🖼️ On-demand involvement in a reply chain](../screenshots/text-generation-on-demand-reply-involvement.webp)


 ### 🪄 Auto Usage
@@ -74,11 +77,14 @@ Prompts may contain the following **placeholder variables** which will be replac
 |---------------------------|-------------|---------|
 | `{{ baibot_name }}`       | Name of the bot as configured in the `user.name` field in the [Static configuration](./README.md#static-configuration) | `Baibot` |
 | `{{ baibot_model_id }}`   | Text-Generation model ID as configured in the [🤖 agent](../agents.md)'s configuration | `gpt-4o` |
-| `{{ baibot_now_utc }}`    | Current date and time in UTC | `2024-09-20 (Friday), 14:26:42 UTC (local timezone/time: unknown)` |
+| `{{ baibot_now_utc }}`    | Current date and time in UTC (⚠️ usage may break prompt caching - see below) | `2024-09-20 (Friday), 14:26:42 UTC` |
+| `{{ baibot_conversation_start_time_utc }}` | The date and time in UTC that the conversation started | `2024-09-20 (Friday), 14:26:42 UTC` |
+
+💡 `{{ baibot_now_utc }}` changes as time goes on, which prevents [prompt caching](https://platform.openai.com/docs/guides/prompt-caching) from working. It's better to use `{{ baibot_conversation_start_time_utc }}` in prompts, as its value doesn't change yet still orients the bot to the current date/time.

 Here's a prompt that combines some of the above variables:

-> You are a brief, but helpful bot called {{ baibot_name }} powered by the {{ baibot_model_id }} model. The date/time now is: {{ baibot_now_utc }}."
+> You are a brief, but helpful bot called {{ baibot_name }} powered by the {{ baibot_model_id }} model. The date/time of this conversation's start is: {{ baibot_conversation_start_time_utc }}."


 ### 🌡️ Temperature Override
--- a/docs/features.md
+++ b/docs/features.md
@@ -28,6 +28,8 @@ Text Generation is the bot's ability to **respond to users' text messages with t

 In multi-user (group) rooms, to avoid disturbing the normal conversation between people, the bot is auto-configured to only respond to messages starting with the command prefix (`!bai`) or direct mentions via the [💬 Text Generation / 🗟 Prefix Requirement Type](./configuration/text-generation.md#-prefix-requirement-type) setting.

+Normally, the bot only responds to allowed [👥 Users](./access.md#-users). In certain cases, it's useful for an allowed user to provoke the bot to respond even in foreign threads or reply chains. You can learn more about this feature in the [On-demand involvement](./features.md#on-demand-involvement) section below.
+
 A few other features (like [🗣️ Text-to-Speech](#️-text-to-speech) and [🦻 Speech-to-Text](#-speech-to-text)) combine well with Text Generation, so you **don't necessarily need to communicate with the bot via text** (with [Seamless voice interaction](#seamless-voice-interaction), you can communicate only with voice).

 You may also wish to see:
@@ -36,6 +38,22 @@ You may also wish to see:
 - [📖 Usage / 💬 Text Generation](./usage.md#-text-generation) section for more details on how to use the bot for Text Generation in a room


+#### On-demand involvement
+
+In the following 2 cases, it's useful to involve the bot in conversations on-demand:
+
+1. In multi-user rooms (with the [🗟 Prefix Requirement](./configuration/text-generation.md#-prefix-requirement-type) setting set to "required")
+2. In rooms with foreign users (users that are not authorized bot [👥 users](./access.md#-users))
+
+In these instances, an allowed [👥 user](./access.md#-users) can also provoke the bot to respond to **any** thread or reply chain by [mentioning](https://spec.matrix.org/latest/client-server-api/#user-and-room-mentions) the bot (e.g. `@baibot Hello!`). The following screenshots demonstrate this behavior:
+
+- [🖼️ On-demand involvement in the room](./screenshots/text-generation-prefix-requirement.webp)
+- [🖼️ On-demand involvement in a thread](./screenshots/text-generation-on-demand-thread-involvement.webp) (the Alice user in this example is not an allowed user, yet her messages are still considered as part of the conversation context)
+- [🖼️ On-demand involvement in a reply chain](./screenshots/text-generation-on-demand-reply-involvement.webp) (the Alice user in this example is not an allowed user, yet her messages are still considered as part of the conversation context)
+
+💡 **NOTE**: Normally, the bot **only considers messages from allowed [👥 Users](./access.md#-users)** and ignores all other messages when responding. However, **when the bot is explicitly invoked (via mention)** in a thread or reply chain, **it will consider all messages** in the thread and reply chain (even those from foreign users) as part of the conversation context.
+
+
 ### 🗣️ Text-to-Speech

 Text-to-Speech is the bot's ability to **turn text messages into voice messages**.
--- a/docs/providers.md
+++ b/docs/providers.md
@@ -125,7 +125,10 @@ For services which are not fully compatible with the OpenAI API, consider using
  - create a room-local agent: `!bai agent create-room-local openai my-openai-agent`
  - create a global agent: `!bai agent create-global openai my-openai-agent`

-💡 When creating an agent, the bot will show you an up-to-date sample configuration for this provider which looks [like this](./sample-provider-configs/openai.yml).
+💡 When creating an agent, the bot will show you an up-to-date sample configuration for this provider which:
+
+- in the general case looks [like this](./sample-provider-configs/openai.yml)
+- for the [o1](https://platform.openai.com/docs/models/o1) models needs to look [like this](./sample-provider-configs/openai-o1.yml)


 ### OpenAI Compatible
--- a/docs/sample-provider-configs/anthropic.yml
+++ b/docs/sample-provider-configs/anthropic.yml
@@ -2,7 +2,7 @@ base_url: https://api.anthropic.com/v1
 api_key: YOUR_API_KEY_HERE
 text_generation:
  model_id: claude-3-5-sonnet-20240620
-  prompt: "You are a brief, but helpful bot called {{ baibot_name }} powered by the {{ baibot_model_id }} model. The date/time now is: {{ baibot_now_utc }}."
+  prompt: "You are a brief, but helpful bot called {{ baibot_name }} powered by the {{ baibot_model_id }} model. The date/time of this conversation's start is: {{ baibot_conversation_start_time_utc }}."
  temperature: 1.0
  max_response_tokens: 8192
  max_context_tokens: 204800
--- a/docs/sample-provider-configs/groq.yml
+++ b/docs/sample-provider-configs/groq.yml
@@ -2,7 +2,7 @@ base_url: https://api.groq.com/openai/v1
 api_key: YOUR_API_KEY_HERE
 text_generation:
  model_id: llama3-70b-8192
-  prompt: "You are a brief, but helpful bot called {{ baibot_name }} powered by the {{ baibot_model_id }} model. The date/time now is: {{ baibot_now_utc }}."
+  prompt: "You are a brief, but helpful bot called {{ baibot_name }} powered by the {{ baibot_model_id }} model. The date/time of this conversation's start is: {{ baibot_conversation_start_time_utc }}."
  temperature: 1.0
  max_response_tokens: 4096
  max_context_tokens: 131072
--- a/docs/sample-provider-configs/localai.yml
+++ b/docs/sample-provider-configs/localai.yml
@@ -2,7 +2,7 @@ base_url: http://my-localai-self-hosted-service:8080/v1
 api_key: YOUR_API_KEY_HERE
 text_generation:
  model_id: gpt-4
-  prompt: "You are a brief, but helpful bot called {{ baibot_name }} powered by the {{ baibot_model_id }} model. The date/time now is: {{ baibot_now_utc }}."
+  prompt: "You are a brief, but helpful bot called {{ baibot_name }} powered by the {{ baibot_model_id }} model. The date/time of this conversation's start is: {{ baibot_conversation_start_time_utc }}."
  temperature: 1.0
  max_response_tokens: 4096
  max_context_tokens: 128000
--- a/docs/sample-provider-configs/mistral.yml
+++ b/docs/sample-provider-configs/mistral.yml
@@ -2,7 +2,7 @@ base_url: https://api.mistral.ai/v1
 api_key: YOUR_API_KEY_HERE
 text_generation:
  model_id: mistral-large-latest
-  prompt: "You are a brief, but helpful bot called {{ baibot_name }} powered by the {{ baibot_model_id }} model. The date/time now is: {{ baibot_now_utc }}."
+  prompt: "You are a brief, but helpful bot called {{ baibot_name }} powered by the {{ baibot_model_id }} model. The date/time of this conversation's start is: {{ baibot_conversation_start_time_utc }}."
  temperature: 1.0
  max_response_tokens: 4096
  max_context_tokens: 128000
--- a/docs/sample-provider-configs/ollama.yml
+++ b/docs/sample-provider-configs/ollama.yml
@@ -2,7 +2,7 @@ base_url: http://my-ollama-self-hosted-service:11434/v1
 api_key: YOUR_API_KEY_HERE
 text_generation:
  model_id: gemma2:2b
-  prompt: "You are a brief, but helpful bot called {{ baibot_name }} powered by the {{ baibot_model_id }} model. The date/time now is: {{ baibot_now_utc }}."
+  prompt: "You are a brief, but helpful bot called {{ baibot_name }} powered by the {{ baibot_model_id }} model. The date/time of this conversation's start is: {{ baibot_conversation_start_time_utc }}."
  temperature: 1.0
  max_response_tokens: 4096
  max_context_tokens: 128000
--- a/docs/sample-provider-configs/openai-compatible.yml
+++ b/docs/sample-provider-configs/openai-compatible.yml
@@ -2,7 +2,7 @@ base_url: ''
 api_key: YOUR_API_KEY_HERE
 text_generation:
  model_id: some-model
-  prompt: "You are a brief, but helpful bot called {{ baibot_name }} powered by the {{ baibot_model_id }} model. The date/time now is: {{ baibot_now_utc }}."
+  prompt: "You are a brief, but helpful bot called {{ baibot_name }} powered by the {{ baibot_model_id }} model. The date/time of this conversation's start is: {{ baibot_conversation_start_time_utc }}."
  temperature: 1.0
  max_response_tokens: 4096
  max_context_tokens: 128000
--- a/docs/sample-provider-configs/openai-o1.yml
+++ b/docs/sample-provider-configs/openai-o1.yml
@@ -0,0 +1,24 @@
+base_url: https://api.openai.com/v1
+api_key: YOUR_API_KEY_HERE
+text_generation:
+  model_id: o1-mini
+  # o1 models do not support a system prompt
+  prompt: null
+  temperature: 1.0
+  # o1 models do not support max_response_tokens.
+  # They use `max_completion_tokens` as an alternative,
+  # but we don't support it yet (see https://github.com/64bit/async-openai/issues/272).
+  max_response_tokens: null
+  max_context_tokens: 128000
+speech_to_text:
+  model_id: whisper-1
+text_to_speech:
+  model_id: tts-1-hd
+  voice: onyx
+  speed: 1.0
+  response_format: opus
+image_generation:
+  model_id: dall-e-3
+  style: vivid
+  size: 1024x1024
+  quality: standard
--- a/docs/sample-provider-configs/openai.yml
+++ b/docs/sample-provider-configs/openai.yml
@@ -1,8 +1,8 @@
 base_url: https://api.openai.com/v1
 api_key: YOUR_API_KEY_HERE
 text_generation:
-  model_id: gpt-4o-2024-08-06
-  prompt: "You are a brief, but helpful bot called {{ baibot_name }} powered by the {{ baibot_model_id }} model. The date/time now is: {{ baibot_now_utc }}."
+  model_id: gpt-4o
+  prompt: "You are a brief, but helpful bot called {{ baibot_name }} powered by the {{ baibot_model_id }} model. The date/time of this conversation's start is: {{ baibot_conversation_start_time_utc }}."
  temperature: 1.0
  max_response_tokens: 16384
  max_context_tokens: 128000
--- a/docs/sample-provider-configs/openrouter.yml
+++ b/docs/sample-provider-configs/openrouter.yml
@@ -2,7 +2,7 @@ base_url: https://openrouter.ai/api/v1
 api_key: YOUR_API_KEY_HERE
 text_generation:
  model_id: mattshumer/reflection-70b:free
-  prompt: "You are a brief, but helpful bot called {{ baibot_name }} powered by the {{ baibot_model_id }} model. The date/time now is: {{ baibot_now_utc }}."
+  prompt: "You are a brief, but helpful bot called {{ baibot_name }} powered by the {{ baibot_model_id }} model. The date/time of this conversation's start is: {{ baibot_conversation_start_time_utc }}."
  temperature: 1.0
  max_response_tokens: 2048
  max_context_tokens: 8192
--- a/docs/sample-provider-configs/together-ai.yml
+++ b/docs/sample-provider-configs/together-ai.yml
@@ -2,7 +2,7 @@ base_url: https://api.together.xyz/v1
 api_key: YOUR_API_KEY_HERE
 text_generation:
  model_id: meta-llama/Meta-Llama-3.1-405B-Instruct-Turbo
-  prompt: "You are a brief, but helpful bot called {{ baibot_name }} powered by the {{ baibot_model_id }} model. The date/time now is: {{ baibot_now_utc }}."
+  prompt: "You are a brief, but helpful bot called {{ baibot_name }} powered by the {{ baibot_model_id }} model. The date/time of this conversation's start is: {{ baibot_conversation_start_time_utc }}."
  temperature: 1.0
  max_response_tokens: 2048
  max_context_tokens: 8192
--- a/docs/screenshots/text-generation-on-demand-reply-involvement.webp
+++ b/docs/screenshots/text-generation-on-demand-reply-involvement.webp
--- a/docs/screenshots/text-generation-on-demand-thread-involvement.webp
+++ b/docs/screenshots/text-generation-on-demand-thread-involvement.webp
--- a/docs/usage.md
+++ b/docs/usage.md
@@ -11,10 +11,11 @@ This is related to the [💬 Text Generation](./features.md#-text-generation) fe

 If there's a text-generation handler agent configured, the bot **may** respond to messages sent in the room.

-🖼️ See screenshots of:
+See screenshots of:

- the [default Text Generation flow](./screenshots/text-generation.webp) for 1:1 rooms
- the [Text Generation flow in multi-user rooms](./screenshots/text-generation-prefix-requirement.webp) (where the [🗟 Prefix Requirement](./configuration/text-generation.md#-prefix-requirement-type) setting is auto-configured to "required")
+- 🖼️ [the default Text Generation flow](./screenshots/text-generation.webp) in 1:1 rooms
+- 🖼️ [the Text Generation flow in multi-user rooms](./screenshots/text-generation-prefix-requirement.webp) (where the [🗟 Prefix Requirement](./configuration/text-generation.md#-prefix-requirement-type) setting is auto-configured to "required")
+- the [on-demand involvement](./features.md#on-demand-involvement) feature

 Whether the bot responds depends on:

@@ -24,9 +25,9 @@ Whether the bot responds depends on:

 - (🎨 agent capabilities) whether the configured `text-generation` (or `catch-all`) handler agent actually supports text-generation. The provider may lack support for this feature or it may be disabled in the [🤖 agents](./agents.md) configuration

- (the [🗟 Prefix Requirement](./configuration/text-generation.md#-prefix-requirement-type) setting) whether a prefix (e.g. `!bai`) is required in front of messages sent to the room. For multi-user rooms, this setting defaults to "required"
+- (the [🗟 Prefix Requirement](./configuration/text-generation.md#-prefix-requirement-type) setting) whether a prefix (e.g. `!bai`) or user mention (e.g. `@baibot`) is required for messages sent to the room. For multi-user rooms, this setting defaults to "required". See [🌟 Features / 💬 Text Generation / On-demand involvement](./features.md#on-demand-involvement) for details.

-Room messages start a threaded conversation where you can continue back-and-forth communication with the bot.
+Room messages start a threaded conversation where you can continue back-and-forth communication with the bot. Using [on-demand involvement](./features.md#on-demand-involvement), you can can also mention the bot to provoke it to get involved in any conversation thread or reply chain.

 Unless you've enabled the [♻️ Context Management](./features.md#️-context-management) feature, all messages will be sent to the agent's API each time. If the context management feature is enabled, older messages may be dropped.

--- a/etc/app/config.yml.dist
+++ b/etc/app/config.yml.dist
@@ -72,8 +72,8 @@ agents:
    #     base_url: https://api.openai.com/v1
    #     api_key: ""
    #     text_generation:
-    #       model_id: gpt-4o-2024-08-06
-    #       prompt: "You are a brief, but helpful bot called {{ baibot_name }} powered by the {{ baibot_model_id }} model. The date/time now is: {{ baibot_now_utc }}."
+    #       model_id: gpt-4o
+    #       prompt: "You are a brief, but helpful bot called {{ baibot_name }} powered by the {{ baibot_model_id }} model. The date/time of this conversation's start is: {{ baibot_conversation_start_time_utc }}."
    #       temperature: 1.0
    #       max_response_tokens: 16384
    #       max_context_tokens: 128000
@@ -97,7 +97,7 @@ agents:
    #     api_key: null
    #     text_generation:
    #       model_id: gpt-4
-    #       prompt: "You are a brief, but helpful bot called {{ baibot_name }} powered by the {{ baibot_model_id }} model. The date/time now is: {{ baibot_now_utc }}."
+    #       prompt: "You are a brief, but helpful bot called {{ baibot_name }} powered by the {{ baibot_model_id }} model. The date/time of this conversation's start is: {{ baibot_conversation_start_time_utc }}."
    #       temperature: 1.0
    #       max_response_tokens: 16384
    #       max_context_tokens: 128000
@@ -122,7 +122,7 @@ agents:
    #     api_key: null
    #     text_generation:
    #       model_id: "gemma2:2b"
-    #       prompt: "You are an assistant based on the gemma2:2b model. Be brief in your responses."
+    #       prompt: "You are a brief, but helpful bot called {{ baibot_name }} powered by the {{ baibot_model_id }} model. The date/time of this conversation's start is: {{ baibot_conversation_start_time_utc }}."
    #       temperature: 1.0
    #       max_response_tokens: 4096
    #       max_context_tokens: 128000
--- a/etc/services/core/compose.yml
+++ b/etc/services/core/compose.yml
@@ -1,6 +1,6 @@
 services:
  postgres:
-    image: docker.io/postgres:16.3-alpine
+    image: docker.io/postgres:16.4-alpine
    user: ${UID}:${GID}
    restart: unless-stopped
    environment:
@@ -13,7 +13,7 @@ services:
    - /etc/passwd:/etc/passwd:ro

  synapse:
-    image: ghcr.io/element-hq/synapse:v1.114.0
+    image: ghcr.io/element-hq/synapse:v1.118.0
    user: "${UID}:${GID}"
    restart: unless-stopped
    entrypoint: python
@@ -26,7 +26,7 @@ services:
    - ./synapse/media-store:/media-store

  element-web:
-    image: docker.io/vectorim/element-web:v1.11.77
+    image: docker.io/vectorim/element-web:v1.11.84
    user: "${UID}:${GID}"
    restart: unless-stopped
    ports:
--- a/etc/services/core/synapse/config/homeserver.yaml
+++ b/etc/services/core/synapse/config/homeserver.yaml
@@ -579,7 +579,9 @@ rc_login:
 #
 #federation_rr_transactions_per_room_per_second: 50

-
+# Authenticated media is not supported yet.
+# See: https://github.com/etkecc/baibot/issues/12
+enable_authenticated_media: false

 # Directory where uploaded images and attachments are stored.
 #
--- a/etc/services/ollama/compose.yml
+++ b/etc/services/ollama/compose.yml
@@ -1,6 +1,6 @@
 services:
  ollama:
-    image: docker.io/ollama/ollama:0.3.9
+    image: docker.io/ollama/ollama:0.4.1
    restart: unless-stopped
    ports:
      - "${SERVICE_OLLAMA_BIND_PORT_HTTP}:11434"
--- a/src/agent/mod.rs
+++ b/src/agent/mod.rs
@@ -21,5 +21,5 @@ pub use provider::{AgentProvider, AgentProviderInfo, ControllerTrait};
 pub use purpose::AgentPurpose;

 pub(super) fn default_prompt() -> &'static str {
-    "You are a brief, but helpful bot called {{ baibot_name }} powered by the {{ baibot_model_id }} model. The date/time now is: {{ baibot_now_utc }}."
+    "You are a brief, but helpful bot called {{ baibot_name }} powered by the {{ baibot_model_id }} model. The date/time of this conversation's start is: {{ baibot_conversation_start_time_utc }}."
 }
--- a/src/agent/provider/anthropic/controller.rs
+++ b/src/agent/provider/anthropic/controller.rs
@@ -2,7 +2,7 @@ use std::fmt::Debug;
 use std::str::FromStr;
 use std::sync::Arc;

-use anthropic_rs::completion::message::ContentType;
+use anthropic_rs::completion::message::{ContentType, System};
 use anthropic_rs::{
    client::Client as AnthropicClient, config::Config as AnthropicConfig,
    models::claude::ClaudeModel,
@@ -72,6 +72,7 @@ impl ControllerTrait for Controller {
        let messages = vec![LLMMessage {
            author: LLMAuthor::User,
            message_text: "Hello!".to_string(),
+            timestamp: chrono::Utc::now(),
        }];

        let conversation = LLMConversation { messages };
@@ -108,6 +109,7 @@ impl ControllerTrait for Controller {
            Some(LLMMessage {
                author: LLMAuthor::Prompt,
                message_text: prompt_text,
+                timestamp: chrono::Utc::now(),
            })
        };

@@ -129,7 +131,7 @@ impl ControllerTrait for Controller {
                &text_generation_config.model_id,
                &prompt_message,
                conversation_messages,
-                text_generation_config.max_response_tokens,
+                Some(text_generation_config.max_response_tokens),
                text_generation_config.max_context_tokens,
            );

@@ -157,7 +159,7 @@ impl ControllerTrait for Controller {
            .unwrap_or(text_generation_config.temperature);

        if let Some(prompt_message) = prompt_message {
-            request.system = Some(prompt_message.message_text);
+            request.system = Some(System::Text(prompt_message.message_text));
        }

        request.model = model;
--- a/src/agent/provider/entity/text_generation/prompt_variables.rs
+++ b/src/agent/provider/entity/text_generation/prompt_variables.rs
@@ -7,17 +7,33 @@ pub struct TextGenerationPromptVariables {

 impl Default for TextGenerationPromptVariables {
    fn default() -> Self {
-        Self::new("unnamed", "unknown-model", Utc::now())
+        let now = Utc::now();
+        Self::new("unnamed", "unknown-model", now, Some(now))
    }
 }

 impl TextGenerationPromptVariables {
-    pub fn new(bot_name: &str, model_id: &str, utc_time: DateTime<Utc>) -> Self {
+    pub fn new(
+        bot_name: &str,
+        model_id: &str,
+        now_time: DateTime<Utc>,
+        conversation_start_time: Option<DateTime<Utc>>,
+    ) -> Self {
        let mut map = HashMap::new();

        map.insert("baibot_name".to_string(), bot_name.to_string());
        map.insert("baibot_model_id".to_string(), model_id.to_string());
-        map.insert("baibot_now_utc".to_string(), format_utc_time(utc_time));
+        map.insert("baibot_now_utc".to_string(), format_utc_time(now_time));
+
+        let baibot_conversation_start_time_utc = match conversation_start_time {
+            Some(conversation_start_time) => format_utc_time(conversation_start_time),
+            None => "unknown".to_string(),
+        };
+
+        map.insert(
+            "baibot_conversation_start_time_utc".to_string(),
+            baibot_conversation_start_time_utc,
+        );

        Self { map }
    }
@@ -52,7 +68,18 @@ mod tests {
            .with_nanosecond(250000000)
            .unwrap();

-        let variables = TextGenerationPromptVariables::new("baibot", "gpt-4o", now_utc);
+        let conversation_start_time_utc = Utc
+            .with_ymd_and_hms(2024, 9, 19, 18, 34, 15)
+            .unwrap()
+            .with_nanosecond(250000000)
+            .unwrap();
+
+        let variables = TextGenerationPromptVariables::new(
+            "baibot",
+            "gpt-4o",
+            now_utc,
+            Some(conversation_start_time_utc),
+        );

        assert_eq!(
            variables.map.get("baibot_name"),
@@ -66,9 +93,13 @@ mod tests {
            variables.map.get("baibot_now_utc"),
            Some(&format_utc_time(now_utc))
        );
+        assert_eq!(
+            variables.map.get("baibot_conversation_start_time_utc"),
+            Some(&format_utc_time(conversation_start_time_utc))
+        );

-        let prompt = "Hello, I'm {{ baibot_name }} using {{ baibot_model_id }}. The date/time now is {{ baibot_now_utc }}.";
-        let expected = "Hello, I'm baibot using gpt-4o. The date/time now is 2024-09-20 (Friday), 18:34:15 UTC.";
+        let prompt = "Hello, I'm {{ baibot_name }} using {{ baibot_model_id }}. The date/time now is {{ baibot_now_utc }} and this conversation started at {{ baibot_conversation_start_time_utc }}.";
+        let expected = "Hello, I'm baibot using gpt-4o. The date/time now is 2024-09-20 (Friday), 18:34:15 UTC and this conversation started at 2024-09-19 (Thursday), 18:34:15 UTC.";

        assert_eq!(variables.format(prompt), expected);
    }
--- a/src/agent/provider/groq/mod.rs
+++ b/src/agent/provider/groq/mod.rs
@@ -15,7 +15,7 @@ pub fn default_config() -> Config {
    if let Some(ref mut config) = config.text_generation.as_mut() {
        config.model_id = "llama3-70b-8192".to_owned();
        config.max_context_tokens = 131_072;
-        config.max_response_tokens = 4096;
+        config.max_response_tokens = Some(4096);
    }

    if let Some(ref mut config) = config.speech_to_text.as_mut() {
--- a/src/agent/provider/localai/mod.rs
+++ b/src/agent/provider/localai/mod.rs
@@ -13,7 +13,7 @@ pub fn default_config() -> Config {
    if let Some(ref mut config) = config.text_generation.as_mut() {
        config.model_id = "gpt-4".to_owned();
        config.max_context_tokens = 128_000;
-        config.max_response_tokens = 4096;
+        config.max_response_tokens = Some(4096);
    }

    if let Some(ref mut config) = config.text_to_speech.as_mut() {
--- a/src/agent/provider/ollama/mod.rs
+++ b/src/agent/provider/ollama/mod.rs
@@ -17,7 +17,7 @@ pub fn default_config() -> Config {
    if let Some(ref mut config) = config.text_generation.as_mut() {
        config.model_id = "gemma2:2b".to_owned();
        config.max_context_tokens = 128_000;
-        config.max_response_tokens = 4096;
+        config.max_response_tokens = Some(4096);
    }

    config
--- a/src/agent/provider/openai/config.rs
+++ b/src/agent/provider/openai/config.rs
@@ -56,7 +56,7 @@ pub struct TextGenerationConfig {
    pub temperature: f32,

    #[serde(default)]
-    pub max_response_tokens: u32,
+    pub max_response_tokens: Option<u32>,

    #[serde(default)]
    pub max_context_tokens: u32,
@@ -68,14 +68,14 @@ impl Default for TextGenerationConfig {
            model_id: default_text_model_id(),
            prompt: Some(default_prompt().to_owned()),
            temperature: super::super::default_temperature(),
-            max_response_tokens: 16_384,
+            max_response_tokens: Some(16_384),
            max_context_tokens: 128_000,
        }
    }
 }

 fn default_text_model_id() -> String {
-    "gpt-4o-2024-08-06".to_owned()
+    "gpt-4o".to_owned()
 }

 #[derive(Debug, Clone, Serialize, Deserialize)]
--- a/src/agent/provider/openai/controller.rs
+++ b/src/agent/provider/openai/controller.rs
@@ -63,6 +63,7 @@ impl ControllerTrait for Controller {
        let messages = vec![LLMMessage {
            author: LLMAuthor::User,
            message_text: "Hello!".to_string(),
+            timestamp: chrono::Utc::now(),
        }];

        let conversation = LLMConversation { messages };
@@ -99,6 +100,7 @@ impl ControllerTrait for Controller {
            Some(LLMMessage {
                author: LLMAuthor::Prompt,
                message_text: prompt_text,
+                timestamp: chrono::Utc::now(),
            })
        };

@@ -131,12 +133,18 @@ impl ControllerTrait for Controller {
            .temperature_override
            .unwrap_or(text_generation_config.temperature);

-        let request = CreateChatCompletionRequestArgs::default()
-            .max_tokens(text_generation_config.max_response_tokens)
+        let mut request_builder = CreateChatCompletionRequestArgs::default();
+
+        request_builder
            .model(&text_generation_config.model_id)
            .temperature(temperature)
-            .messages(openai_conversation_messages)
-            .build()?;
+            .messages(openai_conversation_messages);
+
+        if let Some(max_response_tokens) = text_generation_config.max_response_tokens {
+            request_builder.max_tokens(max_response_tokens);
+        }
+
+        let request = request_builder.build()?;

        if let Ok(request_as_json) = serde_json::to_string(&request) {
            tracing::trace!(
--- a/src/agent/provider/openai_compat/config.rs
+++ b/src/agent/provider/openai_compat/config.rs
@@ -66,7 +66,7 @@ pub struct TextGenerationConfig {
    pub temperature: f32,

    #[serde(default)]
-    pub max_response_tokens: u32,
+    pub max_response_tokens: Option<u32>,

    #[serde(default)]
    pub max_context_tokens: u32,
@@ -78,7 +78,7 @@ impl Default for TextGenerationConfig {
            model_id: default_text_model_id(),
            prompt: Some(default_prompt().to_owned()),
            temperature: super::super::default_temperature(),
-            max_response_tokens: 4096,
+            max_response_tokens: Some(4096),
            max_context_tokens: 128_000,
        }
    }
--- a/src/agent/provider/openai_compat/controller.rs
+++ b/src/agent/provider/openai_compat/controller.rs
@@ -61,6 +61,7 @@ impl ControllerTrait for Controller {
        let messages = vec![LLMMessage {
            author: LLMAuthor::User,
            message_text: "Hello!".to_string(),
+            timestamp: chrono::Utc::now(),
        }];

        let conversation = LLMConversation { messages };
@@ -97,6 +98,7 @@ impl ControllerTrait for Controller {
            Some(LLMMessage {
                author: LLMAuthor::Prompt,
                message_text: prompt_text,
+                timestamp: chrono::Utc::now(),
            })
        };

@@ -131,12 +133,15 @@ impl ControllerTrait for Controller {

        let max_tokens = text_generation_config
            .max_response_tokens
-            .try_into()
-            .expect("Failed converting max_response_tokens from u32 to i32");
+            .map(|max_response_tokens| {
+                max_response_tokens
+                    .try_into()
+                    .expect("Failed converting max_response_tokens from u32 to i32")
+            });

        let request = ChatBody {
            model: text_generation_config.model_id.clone(),
-            max_tokens: Some(max_tokens),
+            max_tokens,
            temperature: Some(temperature),
            top_p: None,
            n: Some(1),
--- a/src/agent/provider/openai_compat/mod.rs
+++ b/src/agent/provider/openai_compat/mod.rs
@@ -56,7 +56,7 @@ pub fn default_config() -> Config {

    if let Some(text_generation) = &mut config.text_generation {
        text_generation.model_id = "some-model".to_string();
-        text_generation.max_response_tokens = 4096;
+        text_generation.max_response_tokens = Some(4096);
        text_generation.max_context_tokens = 128_000;
    }

--- a/src/agent/provider/openrouter/mod.rs
+++ b/src/agent/provider/openrouter/mod.rs
@@ -14,7 +14,7 @@ pub fn default_config() -> Config {
    if let Some(ref mut config) = config.text_generation.as_mut() {
        config.model_id = "mattshumer/reflection-70b:free".to_owned();
        config.max_context_tokens = 8192;
-        config.max_response_tokens = 2048;
+        config.max_response_tokens = Some(2048);
    }

    config
--- a/src/agent/provider/togetherai/mod.rs
+++ b/src/agent/provider/togetherai/mod.rs
@@ -14,7 +14,7 @@ pub fn default_config() -> Config {
    if let Some(ref mut config) = config.text_generation.as_mut() {
        config.model_id = "meta-llama/Meta-Llama-3.1-405B-Instruct-Turbo".to_owned();
        config.max_context_tokens = 8192;
-        config.max_response_tokens = 2048;
+        config.max_response_tokens = Some(2048);
    }

    config
--- a/src/bot/implementation.rs
+++ b/src/bot/implementation.rs
@@ -172,6 +172,24 @@ impl Bot {
        self.matrix_link().user_id()
    }

+    pub(crate) async fn user_display_name_in_room(&self, room: &Room) -> Option<String> {
+        let bot_display_name = self
+            .room_display_name_fetcher()
+            .own_display_name_in_room(room)
+            .await;
+
+        match bot_display_name {
+            Ok(value) => value,
+            Err(err) => {
+                tracing::warn!(
+                    ?err,
+                    "Failed to fetch bot display name. Proceeding without it"
+                );
+                None
+            }
+        }
+    }
+
    pub(crate) fn reacting(&self) -> super::reacting::Reacting {
        super::reacting::Reacting::new(self.clone())
    }
--- a/src/bot/messaging.rs
+++ b/src/bot/messaging.rs
@@ -11,7 +11,7 @@ use mxlink::{CallbackError, MessageResponseType};
 use tracing::Instrument;

 use crate::{
-    conversation::matrix::determine_thread_context_for_room_event,
+    conversation::matrix::determine_interaction_context_for_room_event,
    entity::{MessageContext, MessagePayload, RoomConfigContext, TriggerEventInfo},
 };

@@ -239,8 +239,11 @@ impl Messaging {
            }
        };

-        let thread_context = determine_thread_context_for_room_event(
+        let bot_display_name = self.bot.user_display_name_in_room(&room).await;
+
+        let interaction_context = determine_interaction_context_for_room_event(
            self.bot.user_id(),
+            &bot_display_name,
            &room,
            &event,
            &payload,
@@ -248,16 +251,18 @@ impl Messaging {
        )
        .await;

-        let thread_context = match thread_context {
+        let interaction_context = match interaction_context {
            Ok(value) => value,
            Err(err) => {
-                tracing::error!(?err, "Failed to determine thread context for event");
+                tracing::error!(?err, "Failed to determine interaction context for event");
                return Ok(());
            }
        };

-        let Some(thread_context) = thread_context else {
-            tracing::debug!("Ignoring message with unknown thread context (likely not a threaded message or a top-level message)");
+        let Some(interaction_context) = interaction_context else {
+            tracing::debug!(
+                "Ignoring message with unknown interaction context (likely not a message for us)"
+            );
            return Ok(());
        };

@@ -276,33 +281,14 @@ impl Messaging {
            room_config_context,
            self.bot.admin_pattern_regexes().clone(),
            trigger_event_info,
-            thread_context.info.clone(),
-        );
+            interaction_context.thread_info.clone(),
+        )
+        .with_bot_display_name(bot_display_name);

-        let bot_display_name = self
-            .bot
-            .room_display_name_fetcher()
-            .own_display_name_in_room(message_context.room())
-            .await;
-
-        let bot_display_name = match bot_display_name {
-            Ok(value) => value,
-            Err(err) => {
-                tracing::warn!(
-                    ?err,
-                    "Failed to fetch bot display name. Proceeding without it"
-                );
-                None
-            }
-        };
-
-        // The first event in the thread determines which handler processes the current event.
        let controller_type = crate::controller::determine_controller(
            self.bot.command_prefix(),
-            &thread_context.first_message,
+            &interaction_context.trigger,
            &message_context,
-            self.bot.user_id(),
-            &bot_display_name,
        );

        tracing::info!(?controller_type, "Determined controller");
@@ -310,7 +296,7 @@ impl Messaging {
        let _ = room
            .send_single_receipt(
                ReceiptType::Read,
-                thread_context.info.clone().into(),
+                interaction_context.thread_info.clone().into(),
                event.event_id.clone(),
            )
            .await;
--- a/src/controller/chat_completion/mod.rs
+++ b/src/controller/chat_completion/mod.rs
@@ -18,13 +18,28 @@ use crate::entity::roomconfig::{
 use crate::entity::MessagePayload;
 use crate::strings;
 use crate::utils::text_to_speech::create_transcribed_message_text;
-use crate::{conversation::create_llm_conversation_for_matrix_thread, entity::MessageContext, Bot};
+use crate::{
+    conversation::{
+        create_llm_conversation_for_matrix_reply_chain, create_llm_conversation_for_matrix_thread,
+        matrix::create_list_of_bot_user_prefixes_to_strip,
+    },
+    entity::MessageContext,
+    Bot,
+};

 #[derive(Debug, PartialEq)]
 pub enum ChatCompletionControllerType {
-    ViaText { prefixes_to_strip: Vec<String> },
+    // Invoked via a command prefix (e.g. `!bai Hello!`)
+    TextCommand,
+    // Invoked via a mention (e.g. `@baibot Hello!`)
+    TextMention,
+    // Invoked via a direct message (e.g. `Hello!`)
+    TextDirect,

-    ViaAudio,
+    Audio,
+
+    ThreadMention,
+    ReplyMention,
 }

 struct TextToSpeechEligiblePayload {
@@ -125,7 +140,15 @@ pub async fn handle(
                None
            };

-        let response_type = MessageResponseType::InThread(message_context.thread_info().clone());
+        let response_type = match controller_type {
+            // When we're triggered via a reply mention, we reply to the message that triggered us.
+            ChatCompletionControllerType::ReplyMention => {
+                MessageResponseType::Reply(message_context.thread_info().last_event_id.clone())
+            }
+
+            // In all other cases, we're dealing with a threaded conversation, so we reply in the thread.
+            _ => MessageResponseType::InThread(message_context.thread_info().clone()),
+        };

        let text_to_speech_eligible_payload = handle_stage_text_generation(
            bot,
@@ -353,24 +376,64 @@ async fn handle_stage_text_generation(
    )
    .await?;

-    let prefixes_to_strip = match controller_type {
-        ChatCompletionControllerType::ViaText { prefixes_to_strip } => prefixes_to_strip.clone(),
-        ChatCompletionControllerType::ViaAudio => vec![],
+    // We only strip text from the first message if we're invoked via a command prefix.
+    // Otherwise, we do bot-user mentions stripping on all messages below.
+    let first_message_prefixes_to_strip = match controller_type {
+        ChatCompletionControllerType::TextCommand => vec![bot.command_prefix().to_owned()],
+        _ => vec![],
    };

-    let params = MatrixMessageProcessingParams::new(
-        bot.user_id().as_str().to_owned(),
-        message_context.combined_admin_and_user_regexes(),
-    )
-    .with_first_message_stripped_prefixes(prefixes_to_strip);
+    let bot_user_prefixes_to_strip = create_list_of_bot_user_prefixes_to_strip(
+        bot.user_id(),
+        message_context.bot_display_name(),
+    );

-    let conversation = create_llm_conversation_for_matrix_thread(
-        matrix_link.clone(),
-        message_context.room(),
-        message_context.thread_info().root_event_id.clone(),
-        &params,
-    )
-    .await;
+    let allowed_users = match controller_type {
+        // Regular chat completion only operates on messages from allowed users.
+        ChatCompletionControllerType::TextCommand
+        | ChatCompletionControllerType::TextMention
+        | ChatCompletionControllerType::TextDirect
+        | ChatCompletionControllerType::Audio => {
+            Some(message_context.combined_admin_and_user_regexes())
+        }
+
+        // When we're triggered via an explicit mention (thread or reply), we wish to operate against the mention's whole context
+        // (the whole thread or the whole reply chain upward of the message that triggered us).
+        //
+        // This is to allow admins and users to trigger text-generation for other users' messages.
+        // When we're dragged into a conversation by a known (to us) user, we'd like to process all messages in the conversation,
+        // not just those from allowed users.
+        ChatCompletionControllerType::ThreadMention
+        | ChatCompletionControllerType::ReplyMention => None,
+    };
+
+    let params = MatrixMessageProcessingParams::new(bot.user_id().to_owned(), allowed_users)
+        .with_first_message_prefixes_to_strip(first_message_prefixes_to_strip)
+        .with_bot_user_prefixes_to_strip(bot_user_prefixes_to_strip);
+
+    let conversation = match controller_type {
+        // When we're triggered via a reply mention, the context is the whole reply chain upward of the message that triggered us.
+        ChatCompletionControllerType::ReplyMention => {
+            create_llm_conversation_for_matrix_reply_chain(
+                &bot.room_event_fetcher().clone(),
+                message_context.room(),
+                message_context.thread_info().last_event_id.clone(),
+                &params,
+            )
+            .await
+        }
+
+        // Everything else is happening in a thread, so the context is the whole thread.
+        _ => {
+            create_llm_conversation_for_matrix_thread(
+                matrix_link.clone(),
+                message_context.room(),
+                message_context.thread_info().root_event_id.clone(),
+                &params,
+            )
+            .await
+        }
+    };

    let conversation = match conversation {
        Ok(conversation) => conversation,
@@ -415,6 +478,7 @@ async fn handle_stage_text_generation(
            .text_generation_model_id()
            .unwrap_or("unknown-model".to_owned()),
        chrono::Utc::now(),
+        conversation.start_time(),
    );

    let params = TextGenerationParams {
@@ -548,16 +612,53 @@ async fn handle_stage_speech_to_text_actual_transcribing(
        .instrument(span)
        .await?;

-    let transcribed_text = create_transcribed_message_text(&speech_to_text_result.text);
+    // Only use the `> 🦻 Transcribed text` format if we're posting in a thread.
+    //
+    // If we're dealing with a regular reply (which would be the case in "Transcribe-only mode" = speech-to-text/flow-type=only_transcribe),
+    // we don't want to use the `> 🦻 Transcribed text` format for 2 reasons:
+    //
+    // 1. This kind of blockquote-formatting can be confused by clients for a fallback-for-rich-replies
+    //    (see https://spec.matrix.org/v1.11/client-server-api/#fallbacks-for-rich-replies).
+    //    It makes certain clients render our messages incorrectly.
+    //
+    // 2. Transcribe-only mode is typically used for memos. Sticking to a plain-text format
+    //    allows people to copy-paste the text or forward it to another room more easily (without having to strip formatting, etc.)
+    //
+    // When sending a bare reply, we'd better annotate the message with a 🦻 reaction instead,
+    // to make it clear to users that it's a transcription.
+    //
+    // Regardless of how we post this message, it will be posted as a notice,
+    // which can indicate to the bot (for potential future text-generation purposes) that this message is not a bot message.
+    let (transcribed_text, annotate_message_with_reaction) =
+        if let MessageResponseType::InThread(_) = response_type {
+            (
+                create_transcribed_message_text(&speech_to_text_result.text),
+                false,
+            )
+        } else {
+            (speech_to_text_result.text, true)
+        };

    let result = bot
        .messaging()
        .send_notice_markdown_no_fail(message_context.room(), transcribed_text, response_type)
        .await;

-    result
+    let event_id = result
        .map(|result| result.event_id)
-        .ok_or_else(|| anyhow::anyhow!("Failed to send transcribed text"))
+        .ok_or_else(|| anyhow::anyhow!("Failed to send transcribed text"))?;
+
+    if annotate_message_with_reaction {
+        bot.reacting()
+            .react_no_fail(
+                message_context.room(),
+                event_id.clone(),
+                AgentPurpose::SpeechToText.emoji().to_owned(),
+            )
+            .await;
+    }
+
+    Ok(event_id)
 }

 async fn send_tts_offer_for_message(
--- a/src/controller/determination/mod.rs
+++ b/src/controller/determination/mod.rs
@@ -1,13 +1,11 @@
 #[cfg(test)]
 mod tests;

-use mxlink::matrix_sdk::ruma::OwnedUserId;
-
 use super::chat_completion::ChatCompletionControllerType;
 use crate::{
    entity::{
-        roomconfig::TextGenerationPrefixRequirementType, MessageContext, MessagePayload,
-        ThreadContextFirstMessage,
+        roomconfig::TextGenerationPrefixRequirementType, InteractionTrigger, MessageContext,
+        MessagePayload,
    },
    strings,
 };
@@ -16,12 +14,16 @@ use super::ControllerType;

 pub fn determine_controller(
    command_prefix: &str,
-    first_thread_message: &ThreadContextFirstMessage,
+    first_thread_message: &InteractionTrigger,
    message_context: &MessageContext,
-    bot_user_id: &OwnedUserId,
-    bot_display_name: &Option<String>,
 ) -> ControllerType {
    match &first_thread_message.payload {
+        MessagePayload::SynthethicChatCompletionTriggerInThread => {
+            ControllerType::ChatCompletion(ChatCompletionControllerType::ThreadMention)
+        }
+        MessagePayload::SynthethicChatCompletionTriggerForReply => {
+            ControllerType::ChatCompletion(ChatCompletionControllerType::ReplyMention)
+        }
        MessagePayload::Text(text_message_content) => {
            let prefix_requirement_type = message_context
                .room_config_context()
@@ -32,8 +34,6 @@ pub fn determine_controller(
                &text_message_content.body,
                prefix_requirement_type,
                first_thread_message.is_mentioning_bot,
-                bot_user_id,
-                bot_display_name,
            )
        }
        MessagePayload::Encrypted(thread_info) => {
@@ -47,7 +47,7 @@ pub fn determine_controller(
            }
        }
        MessagePayload::Audio(_) => {
-            ControllerType::ChatCompletion(ChatCompletionControllerType::ViaAudio)
+            ControllerType::ChatCompletion(ChatCompletionControllerType::Audio)
        }
        MessagePayload::Reaction { .. } => {
            panic!("Handling reaction as first message in thread does not make sense")
@@ -60,8 +60,6 @@ fn determine_text_controller(
    text: &str,
    room_text_generation_prefix_requirement_type: TextGenerationPrefixRequirementType,
    is_mentioning_bot: bool,
-    bot_user_id: &OwnedUserId,
-    bot_display_name: &Option<String>,
 ) -> ControllerType {
    let text = text.trim();

@@ -102,53 +100,26 @@ fn determine_text_controller(
    // Otherwise, it depends on the prefix requirement for text generation - it may be routed for chat completion or ignored.

    if is_mentioning_bot {
-        // Different clients do mentions differently.
-        // The body text containing the mention usually contains one of:
-        // - the full user ID (includes a @ prefix by default)
-        // - the localpart (with a @ prefix)
-        // - the localpart (without a @ prefix)
-        // - the display name (with a @ prefix)
-        // - the display name (without a @ prefix)
-        //
-        // Some add a `: ` suffix after the mention.
-        //
-        // There's no guarantee that the mention is at the start even.
-        // It being there is most common and we try to strip it from there
-        // as best as we can.
-        let bot_user_id_localpart = bot_user_id.localpart();
-
-        let mut prefixes_to_strip = vec![
-            bot_user_id.as_str().to_owned(),
-            format!("@{}", bot_user_id_localpart),
-            bot_user_id_localpart.to_owned(),
-        ];
-
-        if let Some(bot_display_name) = bot_display_name {
-            prefixes_to_strip.push(format!("@{}", bot_display_name));
-            prefixes_to_strip.push(bot_display_name.to_owned());
-        }
-
-        prefixes_to_strip.push(":".to_owned());
-
-        return ControllerType::ChatCompletion(ChatCompletionControllerType::ViaText {
-            prefixes_to_strip,
-        });
+        return ControllerType::ChatCompletion(ChatCompletionControllerType::TextMention);
    }

+    // Regardless of what the prefix requirement is, if we encounter a command prefix, we'll consider it a chat completion via command prefix invokation.
+    // This is to correctly indicate to the chat completion controller that a command prefix was used,
+    // so that it can be stripped from the beginning of the message.
+    if text.starts_with(command_prefix) {
+        return ControllerType::ChatCompletion(ChatCompletionControllerType::TextCommand);
+    }
+
+    // We're dealing with a regular message that does not start with a command prefix.
+
    match room_text_generation_prefix_requirement_type {
        TextGenerationPrefixRequirementType::CommandPrefix => {
-            if text.starts_with(command_prefix) {
-                ControllerType::ChatCompletion(ChatCompletionControllerType::ViaText {
-                    prefixes_to_strip: vec![command_prefix.to_owned()],
-                })
-            } else {
-                ControllerType::Ignore
-            }
+            // A prefix is required, but we've already checked (above) that the message does not start with a command prefix.
+            // It's to be ignored.
+            ControllerType::Ignore
        }
        TextGenerationPrefixRequirementType::No => {
-            ControllerType::ChatCompletion(ChatCompletionControllerType::ViaText {
-                prefixes_to_strip: vec![],
-            })
+            ControllerType::ChatCompletion(ChatCompletionControllerType::TextDirect)
        }
    }
 }
--- a/src/controller/determination/tests.rs
+++ b/src/controller/determination/tests.rs
@@ -4,9 +4,6 @@ fn determine_text_controller() {
    use super::ControllerType;
    use crate::controller;

-    let bot_user_id = mxlink::matrix_sdk::ruma::owned_user_id!("@bot:example.com");
-    let bot_display_name = "Bot";
-
    let command_prefix = "!bai";

    struct TestCase {
@@ -44,9 +41,7 @@ fn determine_text_controller() {
            is_mentioning_bot: false,
            room_text_generation_prefix_requirement_type:
                super::TextGenerationPrefixRequirementType::No,
-            expected: ControllerType::ChatCompletion(ChatCompletionControllerType::ViaText {
-                prefixes_to_strip: vec![],
-            }),
+            expected: ControllerType::ChatCompletion(ChatCompletionControllerType::TextCommand),
        },
        TestCase {
            name: "Access top-level",
@@ -110,9 +105,7 @@ fn determine_text_controller() {
            is_mentioning_bot: false,
            room_text_generation_prefix_requirement_type:
                super::TextGenerationPrefixRequirementType::No,
-            expected: ControllerType::ChatCompletion(ChatCompletionControllerType::ViaText {
-                prefixes_to_strip: vec![],
-            }),
+            expected: ControllerType::ChatCompletion(ChatCompletionControllerType::TextDirect),
        },
        TestCase {
            name: "Regular text is ignored when prefix is required",
@@ -128,9 +121,7 @@ fn determine_text_controller() {
            is_mentioning_bot: false,
            room_text_generation_prefix_requirement_type:
                super::TextGenerationPrefixRequirementType::CommandPrefix,
-            expected: ControllerType::ChatCompletion(ChatCompletionControllerType::ViaText {
-                prefixes_to_strip: vec!["!bai".to_owned()],
-            }),
+            expected: ControllerType::ChatCompletion(ChatCompletionControllerType::TextCommand),
        },
        TestCase {
            name: "Command-prefixed text triggers completion even when prefix is not required",
@@ -138,58 +129,35 @@ fn determine_text_controller() {
            is_mentioning_bot: false,
            room_text_generation_prefix_requirement_type:
                super::TextGenerationPrefixRequirementType::No,
-            expected: ControllerType::ChatCompletion(ChatCompletionControllerType::ViaText {
-                prefixes_to_strip: vec![],
-            }),
+            expected: ControllerType::ChatCompletion(ChatCompletionControllerType::TextCommand),
        },
        TestCase {
-            name: "Regular message with bot mention triggers completion stripping bot id and display name (no prefix requirement)",
+            name: "Regular message with bot mention triggers completion (no prefix requirement)",
            input: "Regular text goes here",
            is_mentioning_bot: true,
            room_text_generation_prefix_requirement_type:
                super::TextGenerationPrefixRequirementType::No,
-            expected: ControllerType::ChatCompletion(ChatCompletionControllerType::ViaText {
-                prefixes_to_strip: vec![
-                    "@bot:example.com".to_owned(),
-                    "@bot".to_owned(),
-                    "bot".to_owned(),
-                    "@Bot".to_owned(),
-                    "Bot".to_owned(),
-                    ":".to_owned(),
-                ],
-            }),
+            expected: ControllerType::ChatCompletion(ChatCompletionControllerType::TextMention),
        },
-        // This test case is the same as the one above, just with a different prefix requirement.
+        // This test case is the same as the one above, just with a different prefix requirement setting.
        // We expect the same result.
        TestCase {
-            name: "Regular message with bot mention triggers completion stripping bot id and display name (command_prefix requirement)",
+            name:
+                "Regular message with bot mention triggers completion (command prefix requirement)",
            input: "Regular text goes here",
            is_mentioning_bot: true,
            room_text_generation_prefix_requirement_type:
                super::TextGenerationPrefixRequirementType::CommandPrefix,
-            expected: ControllerType::ChatCompletion(ChatCompletionControllerType::ViaText {
-                prefixes_to_strip: vec![
-                    "@bot:example.com".to_owned(),
-                    "@bot".to_owned(),
-                    "bot".to_owned(),
-                    "@Bot".to_owned(),
-                    "Bot".to_owned(),
-                    ":".to_owned(),
-                ],
-            }),
+            expected: ControllerType::ChatCompletion(ChatCompletionControllerType::TextMention),
        },
    ];

    for test_case in test_cases {
-        let bot_display_name = Some(bot_display_name.to_owned());
-
        let result = super::determine_text_controller(
            command_prefix,
            test_case.input,
            test_case.room_text_generation_prefix_requirement_type,
            test_case.is_mentioning_bot,
-            &bot_user_id,
-            &bot_display_name,
        );
        assert_eq!(result, test_case.expected, "Test case: {}", test_case.name);
    }
--- a/src/controller/image/generation.rs
+++ b/src/controller/image/generation.rs
@@ -35,8 +35,8 @@ pub async fn handle_image(
    };

    let params = MatrixMessageProcessingParams::new(
-        bot.user_id().as_str().to_owned(),
-        message_context.combined_admin_and_user_regexes(),
+        bot.user_id().to_owned(),
+        Some(message_context.combined_admin_and_user_regexes()),
    );

    let conversation = create_llm_conversation_for_matrix_thread(
--- a/src/controller/image/prompt.rs
+++ b/src/controller/image/prompt.rs
@@ -46,6 +46,8 @@ mod tests {

    #[test]
    fn test_build_prompt() {
+        let timestamp = chrono::Utc::now();
+
        let test_cases = vec![
            // Simple case
            TestCase {
@@ -59,6 +61,7 @@ mod tests {
                messages: vec![Message {
                    author: Author::User,
                    message_text: "Must be blue".to_owned(),
+                    timestamp,
                }],
                expected_prompt: "Generate a picture of a dog\nOther criteria:\n- Must be blue",
            },
@@ -68,14 +71,17 @@ mod tests {
                messages: vec![Message {
                    author: Author::User,
                    message_text: "Must be blue".to_owned(),
+                    timestamp,
                },
                Message {
                    author: Author::Assistant,
                    message_text: "Whatever".to_owned(),
+                    timestamp,
                },
                Message {
                    author: Author::User,
                    message_text: "Must be 3-legged.\nMust be flying.".to_owned(),
+                    timestamp,
                }],
                expected_prompt: "Generate a picture of an elephant\nOther criteria:\n- Must be blue\n- Must be 3-legged.. Must be flying.",
            },
@@ -85,18 +91,22 @@ mod tests {
                messages: vec![Message {
                    author: Author::User,
                    message_text: "Must be blue".to_owned(),
+                    timestamp,
                },
                Message {
                    author: Author::Assistant,
                    message_text: "Whatever".to_owned(),
+                    timestamp,
                },
                Message {
                    author: Author::User,
                    message_text: "Again".to_owned(),
+                    timestamp,
                },
                Message {
                    author: Author::User,
                    message_text: "again".to_owned(),
+                    timestamp,
                }],
                expected_prompt: "Generate a picture of a grizzly bear\nOther criteria:\n- Must be blue",
            },
--- a/src/conversation/llm/entity.rs
+++ b/src/conversation/llm/entity.rs
@@ -1,3 +1,5 @@
+use chrono::{DateTime, Utc};
+
 #[derive(Debug, Clone, PartialEq)]
 pub enum Author {
    Prompt,
@@ -9,6 +11,7 @@ pub enum Author {
 pub struct Message {
    pub author: Author,
    pub message_text: String,
+    pub timestamp: DateTime<Utc>,
 }

 #[derive(Debug)]
@@ -52,39 +55,59 @@ impl Conversation {
            messages: new_messages,
        }
    }
+
+    pub fn start_time(&self) -> Option<DateTime<Utc>> {
+        self.messages.first().map(|message| message.timestamp)
+    }
 }

 #[cfg(test)]
 mod tests {
    use super::*;
+    use chrono::{TimeZone, Utc};

    #[test]
    fn combine_consecutive_messages() {
+        let timestamp_1 = Utc.with_ymd_and_hms(2024, 9, 20, 18, 34, 15).unwrap();
+
+        let timestamp_2 = Utc.with_ymd_and_hms(2024, 9, 21, 18, 34, 15).unwrap();
+
+        let timestamp_3 = Utc.with_ymd_and_hms(2024, 9, 22, 18, 34, 15).unwrap();
+
        let conversation = Conversation {
            messages: vec![
+                // User's turn
                Message {
                    author: Author::User,
                    message_text: "Hello".to_string(),
+                    timestamp: timestamp_1,
                },
                Message {
                    author: Author::User,
                    message_text: "How are you?".to_string(),
+                    timestamp: timestamp_2,
                },
                Message {
                    author: Author::User,
                    message_text: "I'm OK, btw.".to_string(),
+                    timestamp: timestamp_3,
                },
+                // Assistant's turn
                Message {
                    author: Author::Assistant,
                    message_text: "Hi there!".to_string(),
+                    timestamp: timestamp_2,
                },
                Message {
                    author: Author::Assistant,
                    message_text: "I'm doing well, thank you.".to_string(),
+                    timestamp: timestamp_3,
                },
+                // User's turn
                Message {
                    author: Author::User,
                    message_text: "That's great!".to_string(),
+                    timestamp: timestamp_3,
                },
            ],
        };
@@ -97,12 +120,17 @@ mod tests {
            conversation.messages[0].message_text,
            "Hello\nHow are you?\nI'm OK, btw."
        );
+        assert_eq!(conversation.messages[0].timestamp, timestamp_1);
+
        assert_eq!(conversation.messages[1].author, Author::Assistant);
        assert_eq!(
            conversation.messages[1].message_text,
            "Hi there!\nI'm doing well, thank you."
        );
+        assert_eq!(conversation.messages[1].timestamp, timestamp_2);
+
        assert_eq!(conversation.messages[2].author, Author::User);
        assert_eq!(conversation.messages[2].message_text, "That's great!");
+        assert_eq!(conversation.messages[2].timestamp, timestamp_3);
    }
 }
--- a/src/conversation/llm/tests.rs
+++ b/src/conversation/llm/tests.rs
@@ -1,3 +1,5 @@
+use mxlink::matrix_sdk::ruma::OwnedUserId;
+
 use crate::utils::status::create_error_message_text;
 use crate::utils::text_to_speech::create_transcribed_message_text;

@@ -5,15 +7,18 @@ use super::*;

 #[test]
 fn test_messages_by_the_bot_are_identified_correctly() {
-    let bot_user_id = "@bot:example.com";
+    let bot_user_id =
+        OwnedUserId::try_from("@bot:example.com").expect("Failed to parse bot user ID");

    let matrix_message = super::super::matrix::MatrixMessage {
        sender_id: bot_user_id.to_owned(),
        message_type: super::super::matrix::MatrixMessageType::Text,
        message_text: "Hello!".to_owned(),
+        mentioned_users: vec![],
+        timestamp: chrono::Utc::now(),
    };

-    let llm_message = convert_matrix_message_to_llm_message(&matrix_message, bot_user_id).unwrap();
+    let llm_message = convert_matrix_message_to_llm_message(&matrix_message, &bot_user_id).unwrap();

    assert_eq!(llm_message.author, Author::Assistant);
    assert_eq!(llm_message.message_text, "Hello!");
@@ -22,7 +27,8 @@ fn test_messages_by_the_bot_are_identified_correctly() {
 #[test]
 fn test_notice_messages_by_bot_with_speech_to_text_prefix_are_cleaned_up_and_considered_sent_by_user(
 ) {
-    let bot_user_id = "@bot:example.com";
+    let bot_user_id =
+        OwnedUserId::try_from("@bot:example.com").expect("Failed to parse bot user ID");

    let source_message_text = "Hello!";
    let message_text = create_transcribed_message_text(source_message_text);
@@ -33,9 +39,11 @@ fn test_notice_messages_by_bot_with_speech_to_text_prefix_are_cleaned_up_and_con
        sender_id: bot_user_id.to_owned(),
        message_type: super::super::matrix::MatrixMessageType::Notice,
        message_text,
+        mentioned_users: vec![],
+        timestamp: chrono::Utc::now(),
    };

-    let llm_message = convert_matrix_message_to_llm_message(&matrix_message, bot_user_id).unwrap();
+    let llm_message = convert_matrix_message_to_llm_message(&matrix_message, &bot_user_id).unwrap();

    assert_eq!(llm_message.author, Author::User);
    assert_eq!(llm_message.message_text, source_message_text);
@@ -43,7 +51,8 @@ fn test_notice_messages_by_bot_with_speech_to_text_prefix_are_cleaned_up_and_con

 #[test]
 fn test_notice_error_messages_by_bot_are_ignored() {
-    let bot_user_id = "@bot:example.com";
+    let bot_user_id =
+        OwnedUserId::try_from("@bot:example.com").expect("Failed to parse bot user ID");

    let source_message_text = "Some error happened";
    let message_text = create_error_message_text(source_message_text);
@@ -54,9 +63,11 @@ fn test_notice_error_messages_by_bot_are_ignored() {
        sender_id: bot_user_id.to_owned(),
        message_type: super::super::matrix::MatrixMessageType::Notice,
        message_text,
+        mentioned_users: vec![],
+        timestamp: chrono::Utc::now(),
    };

-    let llm_message = convert_matrix_message_to_llm_message(&matrix_message, bot_user_id);
+    let llm_message = convert_matrix_message_to_llm_message(&matrix_message, &bot_user_id);

    assert!(llm_message.is_none());
 }
@@ -68,7 +79,8 @@ fn test_other_notice_messages_by_the_bot_are_ignored() {
    // (except for speech-to-text-created transcriptions - see `test_notice_messages_by_bot_with_speech_to_text_prefix_are_cleaned_up_and_considered_sent_by_user()`).
    // This test is to make sure that we don't accidentally start accepting other notice messages.

-    let bot_user_id = "@bot:example.com";
+    let bot_user_id =
+        OwnedUserId::try_from("@bot:example.com").expect("Failed to parse bot user ID");

    let message_text = "Something something";

@@ -76,9 +88,11 @@ fn test_other_notice_messages_by_the_bot_are_ignored() {
        sender_id: bot_user_id.to_owned(),
        message_type: super::super::matrix::MatrixMessageType::Notice,
        message_text: message_text.to_owned(),
+        mentioned_users: vec![],
+        timestamp: chrono::Utc::now(),
    };

-    let llm_message = convert_matrix_message_to_llm_message(&matrix_message, bot_user_id);
+    let llm_message = convert_matrix_message_to_llm_message(&matrix_message, &bot_user_id);

    assert!(llm_message.is_none());
 }
--- a/src/conversation/llm/tokenization.rs
+++ b/src/conversation/llm/tokenization.rs
@@ -16,7 +16,7 @@ pub fn shorten_messages_list_to_context_size(
    model: &str,
    prompt_message: &Option<Message>,
    mut messages: Vec<Message>,
-    max_response_tokens: u32,
+    max_response_tokens: Option<u32>,
    max_context_tokens: u32,
 ) -> Vec<Message> {
    // Loading the tokenization data is an expensive process, so
@@ -26,7 +26,8 @@ pub fn shorten_messages_list_to_context_size(
    // We want to retain the prompt in all cases, so we always count it first.
    // We also always reserve enough tokens for the maximum response we expect.
    let mut current_context_length: u32 = if let Some(prompt_message) = prompt_message {
-        calculate_token_size_for_message(&bpe, model, prompt_message) + max_response_tokens
+        calculate_token_size_for_message(&bpe, model, prompt_message)
+            + max_response_tokens.unwrap_or(0)
    } else {
        0
    };
@@ -85,6 +86,7 @@ pub mod test {
        let message = super::Message {
            author: super::Author::User,
            message_text: "Hello there!".to_owned(),
+            timestamp: chrono::Utc::now(),
        };

        let tokens = super::calculate_token_size_for_message(&bpe, model, &message);
@@ -98,11 +100,12 @@ pub mod test {

        let bpe = super::get_bpe_for_model(model);

-        let max_response_tokens: u32 = 5;
+        let max_response_tokens: Option<u32> = Some(5);

        let prompt = super::Message {
            author: super::Author::Prompt,
            message_text: "You are a bot!".to_owned(),
+            timestamp: chrono::Utc::now(),
        };
        let prompt_length = 10;

@@ -116,6 +119,7 @@ pub mod test {
        let first = super::Message {
            author: super::Author::User,
            message_text: "Hello there!".to_owned(),
+            timestamp: chrono::Utc::now(),
        };
        let first_length = 8;

@@ -129,6 +133,7 @@ pub mod test {
        let second = super::Message {
            author: super::Author::Assistant,
            message_text: "Hello!".to_owned(),
+            timestamp: chrono::Utc::now(),
        };
        let second_length = 7;

@@ -143,6 +148,7 @@ pub mod test {
            author: super::Author::User,
            message_text: "This is the 3rd message in this conversation. It shall be preserved."
                .to_owned(),
+            timestamp: chrono::Utc::now(),
        };
        let third_length = 21;

@@ -156,6 +162,7 @@ pub mod test {
        let forth = super::Message {
            author: super::Author::Assistant,
            message_text: "This is yet another message that shall be preserved.".to_owned(),
+            timestamp: chrono::Utc::now(),
        };
        let forth_length = 15;

@@ -173,7 +180,7 @@ pub mod test {
            &Some(prompt),
            conversation_messages,
            max_response_tokens,
-            prompt_length + max_response_tokens + forth_length + third_length,
+            prompt_length + max_response_tokens.unwrap_or(0) + forth_length + third_length,
        );

        assert_eq!(2, new_conversation_messages.len());
@@ -195,11 +202,12 @@ pub mod test {

        let bpe = super::get_bpe_for_model(model);

-        let max_response_tokens: u32 = 5;
+        let max_response_tokens: Option<u32> = Some(5);

        let prompt = super::Message {
            author: super::Author::User,
            message_text: "あなたはボットです。".to_owned(),
+            timestamp: chrono::Utc::now(),
        };
        let prompt_length = 14;

@@ -213,6 +221,7 @@ pub mod test {
        let first = super::Message {
            author: super::Author::User,
            message_text: "こんにちは!".to_owned(),
+            timestamp: chrono::Utc::now(),
        };
        let first_length = 7;

@@ -226,6 +235,7 @@ pub mod test {
        let second = super::Message {
            author: super::Author::Assistant,
            message_text: "こんにちは。今日は元気ですか。".to_owned(),
+            timestamp: chrono::Utc::now(),
        };
        let second_length = 15;

@@ -239,6 +249,7 @@ pub mod test {
        let third = super::Message {
            author: super::Author::User,
            message_text: "これは第3のメッセージなので、保存されます。".to_owned(),
+            timestamp: chrono::Utc::now(),
        };
        let third_length = 22;

@@ -252,6 +263,7 @@ pub mod test {
        let forth = super::Message {
            author: super::Author::Assistant,
            message_text: "これはもう一つの保存されますメッセージです。".to_owned(),
+            timestamp: chrono::Utc::now(),
        };
        let forth_length = 21;

@@ -269,7 +281,7 @@ pub mod test {
            &Some(prompt),
            conversation_messages,
            max_response_tokens,
-            prompt_length + max_response_tokens + forth_length + third_length,
+            prompt_length + max_response_tokens.unwrap_or(0) + forth_length + third_length,
        );

        assert_eq!(2, new_conversation_messages.len());
--- a/src/conversation/llm/utils.rs
+++ b/src/conversation/llm/utils.rs
@@ -1,12 +1,14 @@
+use matrix_sdk::ruma::OwnedUserId;
+
 use super::{Author, Message};
 use crate::conversation::matrix::{MatrixMessage, MatrixMessageType};
 use crate::utils::text_to_speech as text_to_speech_utils;

 pub fn convert_matrix_message_to_llm_message(
    matrix_message: &MatrixMessage,
-    bot_user_id: &str,
+    bot_user_id: &OwnedUserId,
 ) -> Option<Message> {
-    if matrix_message.sender_id == bot_user_id {
+    if matrix_message.sender_id == bot_user_id.as_str() {
        return convert_bot_message(matrix_message);
    }

@@ -15,28 +17,43 @@ pub fn convert_matrix_message_to_llm_message(

 fn convert_bot_message(matrix_message: &MatrixMessage) -> Option<Message> {
    match matrix_message.message_type {
-        MatrixMessageType::Text => convert_bot_text_message(&matrix_message.message_text),
-        MatrixMessageType::Notice => convert_bot_notice_message(&matrix_message.message_text),
+        MatrixMessageType::Text => {
+            convert_bot_text_message(&matrix_message.message_text, &matrix_message.timestamp)
+        }
+        MatrixMessageType::Notice => {
+            convert_bot_notice_message(&matrix_message.message_text, &matrix_message.timestamp)
+        }
    }
 }

-fn convert_bot_text_message(text: &str) -> Option<Message> {
+fn convert_bot_text_message(
+    text: &str,
+    timestamp: &chrono::DateTime<chrono::Utc>,
+) -> Option<Message> {
    Some(Message {
        author: Author::Assistant,
        message_text: text.to_owned(),
+        timestamp: timestamp.to_owned(),
    })
 }

-fn convert_bot_notice_message(text: &str) -> Option<Message> {
+fn convert_bot_notice_message(
+    text: &str,
+    timestamp: &chrono::DateTime<chrono::Utc>,
+) -> Option<Message> {
    // Notice messages sent by the bot are usually transcriptions of previous messages sent by the user.
    // Such transcriptions are prefixed with an emoji and blockquoted.
    // If we find a notice that doesn't match this pattern, we skip it.
+    //
+    // It should be noted that transcriptions are sometimes posted as regular notice messages which do not include
+    // the `> 🦻` formatting. This function will not handle these properly.

    if let Some(text) = text_to_speech_utils::parse_transcribed_message_text(text) {
        // This is a transcription message. We remove the prefix and consider it as a message sent by the user.
        return Some(Message {
            author: Author::User,
            message_text: text.to_owned(),
+            timestamp: timestamp.to_owned(),
        });
    }

@@ -47,5 +64,6 @@ fn convert_user_message(matrix_message: &MatrixMessage) -> Option<Message> {
    Some(Message {
        author: Author::User,
        message_text: matrix_message.message_text.clone(),
+        timestamp: matrix_message.timestamp.to_owned(),
    })
 }
--- a/src/conversation/matrix/entity.rs
+++ b/src/conversation/matrix/entity.rs
@@ -1,10 +1,15 @@
+use chrono::{DateTime, Utc};
 use regex::Regex;

+use mxlink::matrix_sdk::ruma::OwnedUserId;
+
 #[derive(Clone)]
 pub struct MatrixMessage {
-    pub sender_id: String,
+    pub sender_id: OwnedUserId,
    pub message_type: MatrixMessageType,
    pub message_text: String,
+    pub mentioned_users: Vec<OwnedUserId>,
+    pub timestamp: DateTime<Utc>,
 }

 #[derive(Clone)]
@@ -13,26 +18,42 @@ pub enum MatrixMessageType {
    Notice,
 }

-#[derive(Default, Clone)]
+#[derive(Clone)]
 pub struct MatrixMessageProcessingParams {
-    pub(crate) bot_user_id: String,
-    pub(crate) allowed_users: Vec<Regex>,
+    pub(crate) bot_user_id: OwnedUserId,

-    // If non-empty, these prefixes will be stripped when processing the message
-    pub(crate) first_message_stripped_prefixes: Vec<String>,
+    /// The prefixes that will be stripped when processing the messages in the context (thread or reply chain),
+    /// which are found to be mentioning the bot user (`bot_user_id`).
+    pub(crate) bot_user_prefixes_to_strip: Vec<String>,
+
+    /// The prefixes that will be stripped when processing the 1st message in the context (thread or reply chain).
+    pub(crate) first_message_prefixes_to_strip: Vec<String>,
+
+    /// A list of users whose messages are allowed.
+    /// If None, all messages are allowed.
+    /// If Some, only messages from the allowed users (and the bot itself, `bot_user_id`) are allowed.
+    pub(crate) allowed_users: Option<Vec<Regex>>,
 }

 impl MatrixMessageProcessingParams {
-    pub fn new(bot_user_id: String, allowed_users: Vec<Regex>) -> Self {
+    pub fn new(bot_user_id: OwnedUserId, allowed_users: Option<Vec<Regex>>) -> Self {
        Self {
            bot_user_id,
+            bot_user_prefixes_to_strip: vec![],
+
+            first_message_prefixes_to_strip: vec![],
+
            allowed_users,
-            ..Default::default()
        }
    }

-    pub fn with_first_message_stripped_prefixes(mut self, value: Vec<String>) -> Self {
-        self.first_message_stripped_prefixes = value;
+    pub fn with_bot_user_prefixes_to_strip(mut self, value: Vec<String>) -> Self {
+        self.bot_user_prefixes_to_strip = value;
+        self
+    }
+
+    pub fn with_first_message_prefixes_to_strip(mut self, value: Vec<String>) -> Self {
+        self.first_message_prefixes_to_strip = value;
        self
    }
 }
--- a/src/conversation/matrix/utils/mod.rs
+++ b/src/conversation/matrix/utils/mod.rs
@@ -5,9 +5,12 @@ use std::sync::Arc;

 use mxlink::matrix_sdk::ruma::{OwnedEventId, OwnedUserId};
 use mxlink::matrix_sdk::{
+    deserialized_responses::TimelineEvent,
    ruma::events::{
+        relation::Thread,
        room::message::{
-            MessageType, OriginalSyncRoomMessageEvent, Relation, RoomMessageEventContent,
+            sanitize::remove_plain_reply_fallback, MessageType, OriginalSyncRoomMessageEvent,
+            Relation, RoomMessageEventContent,
        },
        AnyMessageLikeEvent, AnyMessageLikeEventContent, AnyTimelineEvent, MessageLikeEvent,
    },
@@ -16,7 +19,12 @@ use mxlink::matrix_sdk::{
 use mxlink::{MatrixLink, ThreadGetMessagesParams, ThreadInfo};

 use super::{MatrixMessage, MatrixMessageProcessingParams, MatrixMessageType, RoomEventFetcher};
-use crate::entity::{MessagePayload, ThreadContext, ThreadContextFirstMessage};
+use crate::entity::{InteractionContext, InteractionTrigger, MessagePayload};
+
+struct DetailedMessagePayload {
+    is_mentioning_bot: bool,
+    message_payload: MessagePayload,
+}

 pub async fn get_matrix_messages_in_thread(
    matrix_link: MatrixLink,
@@ -42,23 +50,123 @@ pub async fn get_matrix_messages_in_thread(
    Ok(messages)
 }

-pub async fn process_matrix_messages_in_thread(
+pub async fn get_matrix_messages_in_reply_chain(
+    event_fetcher: &Arc<RoomEventFetcher>,
+    room: &Room,
+    event_id: OwnedEventId,
+) -> Result<Vec<MatrixMessage>, mxlink::matrix_sdk::Error> {
+    let messages_native =
+        get_matrix_messages_in_reply_chain_native(event_fetcher, room, event_id).await?;
+
+    let mut messages: Vec<MatrixMessage> = Vec::new();
+
+    for matrix_native_message in messages_native {
+        let Some(message) = convert_matrix_native_event_to_matrix_message(&matrix_native_message)
+        else {
+            continue;
+        };
+
+        messages.push(message);
+    }
+
+    Ok(messages)
+}
+
+async fn get_matrix_messages_in_reply_chain_native(
+    event_fetcher: &Arc<RoomEventFetcher>,
+    room: &Room,
+    event_id: OwnedEventId,
+) -> Result<Vec<AnyMessageLikeEvent>, mxlink::matrix_sdk::Error> {
+    let mut next_event_id = Some(event_id.clone());
+
+    let mut messages: Vec<AnyMessageLikeEvent> = Vec::new();
+    let mut handled_event_ids: Vec<OwnedEventId> = Vec::new();
+
+    while let Some(next_event_id_in_loop) = next_event_id {
+        let event = event_fetcher
+            .fetch_event_in_room(&next_event_id_in_loop, room)
+            .await
+            .unwrap();
+
+        if handled_event_ids.contains(&next_event_id_in_loop) {
+            tracing::warn!(
+                "Not following loop-causing event: {}",
+                next_event_id_in_loop
+            );
+            break;
+        }
+
+        handled_event_ids.push(next_event_id_in_loop.clone());
+
+        let event_deserialized = event.event.deserialize()?;
+
+        let AnyTimelineEvent::MessageLike(message_like_event) = event_deserialized else {
+            tracing::warn!(
+                "Not proceeding past non-MessageLike event: {:?}",
+                event_deserialized
+            );
+            break;
+        };
+
+        next_event_id = match message_like_event.clone() {
+            AnyMessageLikeEvent::RoomEncrypted(_) => None,
+            AnyMessageLikeEvent::RoomMessage(room_message) => {
+                if let MessageLikeEvent::Original(room_message_original) = room_message {
+                    match room_message_original.content.relates_to {
+                        Some(Relation::Reply { in_reply_to }) => Some(in_reply_to.event_id.clone()),
+                        _ => None,
+                    }
+                } else {
+                    None
+                }
+            }
+            _ => None,
+        };
+
+        messages.push(message_like_event);
+    }
+
+    messages.reverse();
+
+    Ok(messages)
+}
+
+pub async fn process_matrix_messages(
    messages: &[MatrixMessage],
    params: &MatrixMessageProcessingParams,
 ) -> Vec<MatrixMessage> {
    let mut messages_filtered: Vec<MatrixMessage> = Vec::new();

    for (i, message) in messages.iter().enumerate() {
-        if !is_message_from_allowed_sender(message, &params.bot_user_id, &params.allowed_users) {
+        if !is_message_from_allowed_sender(
+            message,
+            &params.bot_user_id,
+            params.allowed_users.as_deref(),
+        ) {
            continue;
        }

        let mut message = message.clone();

-        if i == 0 && !params.first_message_stripped_prefixes.is_empty() {
+        if i == 0 && !params.first_message_prefixes_to_strip.is_empty() {
            let mut message_text = message.message_text.clone();

-            for prefix in &params.first_message_stripped_prefixes {
+            for prefix in &params.first_message_prefixes_to_strip {
+                if let Some(message_text_stripped) = message_text.strip_prefix(prefix) {
+                    message_text = message_text_stripped.to_owned();
+                }
+            }
+
+            message.message_text = message_text.trim().to_owned();
+        }
+
+        // We only strip `bot_user_prefixes_to_strip`-defined prefixes from messages that mention the bot user.
+        if !params.bot_user_prefixes_to_strip.is_empty()
+            && message.mentioned_users.contains(&params.bot_user_id)
+        {
+            let mut message_text = message.message_text.clone();
+
+            for prefix in &params.bot_user_prefixes_to_strip {
                if let Some(message_text_stripped) = message_text.strip_prefix(prefix) {
                    message_text = message_text_stripped.to_owned();
                }
@@ -73,16 +181,25 @@ pub async fn process_matrix_messages_in_thread(
    messages_filtered
 }

+/// Tells if the given message is from an allowed sender.
+///
+/// If allowed_users is None, all messages are allowed.
+/// If allowed_users is Some, only messages from the allowed users (and the `bot_user_id`) are allowed.
 fn is_message_from_allowed_sender(
    matrix_message: &MatrixMessage,
-    bot_user_id: &str,
-    allowed_users: &[regex::Regex],
+    bot_user_id: &OwnedUserId,
+    allowed_users: Option<&[regex::Regex]>,
 ) -> bool {
-    if matrix_message.sender_id == bot_user_id {
+    if matrix_message.sender_id == *bot_user_id {
        return true;
    }

-    if mxidwc::match_user_id(&matrix_message.sender_id, allowed_users) {
+    if let Some(allowed_users) = allowed_users {
+        if mxidwc::match_user_id(matrix_message.sender_id.as_str(), allowed_users) {
+            return true;
+        }
+    } else {
+        // No allowed users configured, so all messages are allowed
        return true;
    }

@@ -108,30 +225,67 @@ pub fn convert_matrix_native_event_to_matrix_message(
        _ => return None,
    };

+    let is_reply = matches!(room_message.relates_to, Some(Relation::Reply { .. }));
+
+    let text = if is_reply {
+        // For regular replies, we need to strip the fallback-for-rich replies part.
+        // See: https://spec.matrix.org/v1.11/client-server-api/#fallbacks-for-rich-replies
+        remove_plain_reply_fallback(&text).to_owned()
+    } else {
+        text
+    };
+
+    let timestamp = chrono::DateTime::<chrono::Utc>::from(
+        matrix_native_event
+            .origin_server_ts()
+            .to_system_time()
+            .unwrap_or_else(std::time::SystemTime::now),
+    );
+
+    let mentioned_users = room_message
+        .mentions
+        .map(|m| m.user_ids.iter().map(|u| u.to_owned()).collect())
+        .unwrap_or(vec![]);
+
    Some(MatrixMessage {
-        sender_id: matrix_native_event.sender().to_string(),
+        sender_id: matrix_native_event.sender().to_owned(),
        message_type: if is_notice {
            MatrixMessageType::Notice
        } else {
            MatrixMessageType::Text
        },
        message_text: text,
+        mentioned_users,
+        timestamp,
    })
 }

-/// Determines the thread context (relationship within the thread + first thread message payload) for an incoming (new) room event.
-/// This room event is assumed to be the "newest message" in the thread (or a top-level message).
-/// If the given event is a regular reply (not a thread reply), this function will return `None`.
-/// If the given event is a top-level message, this function will consider this event as the start of the thread.
-/// If the given event is a thread reply, this function will inspect the thread root event and will return the thread context.
-/// If the thread root event is not found, is redacted, or is of some unsupported MessagePayload type, this function will return `None`.
-pub async fn determine_thread_context_for_room_event(
+/// Determines the interaction context for an incoming (new) room event.
+///
+/// This context is created based on the "newest message" (`current_event`), which is:
+/// - either a top-level message, which may or may not be mentioning the bot
+///     - this function will inspect the event and will likely start a new threaded conversation
+///
+/// - or a thread reply
+///     - this function will inspect the thread root event and will return the interaction context
+///     - if the bot only reacts to prefixed messsages (or mentions), this function may ignore the given thread reply, unless it mentions the bot (which causes a synthetic "first message" to be produced)
+///     - if the thread root event is not found, is redacted, or is of some unsupported MessagePayload type, this function will return `None`
+///
+/// - or an in-room (non-threaded) reply to a room message, which may or may not be mentioning the bot
+///     - replies that do not mention the bot cause this function to return `None`
+///     - other replies create a interaction context which points to a "first message" which is synthetic
+#[tracing::instrument(name = "determine_interaction_context_for_room_event", skip_all, fields(room_id = room.room_id().as_str(), event_id = current_event.event_id.as_str()))]
+pub async fn determine_interaction_context_for_room_event(
    bot_user_id: &OwnedUserId,
+    bot_display_name: &Option<String>,
    room: &Room,
    current_event: &OriginalSyncRoomMessageEvent,
    current_event_payload: &MessagePayload,
    event_fetcher: &Arc<RoomEventFetcher>,
-) -> anyhow::Result<Option<ThreadContext>> {
+) -> anyhow::Result<Option<InteractionContext>> {
+    let current_event_is_mentioning_bot =
+        is_event_mentioning_bot(&current_event.content, bot_user_id, bot_display_name);
+
    let Some(relation) = &current_event.content.relates_to else {
        // This is a top-level message. We consider it the start of the thread.
        let thread_info = ThreadInfo::new(
@@ -139,25 +293,75 @@ pub async fn determine_thread_context_for_room_event(
            current_event.event_id.clone(),
        );

-        let is_mentioning_bot = is_event_mentioning_bot(&current_event.content, bot_user_id);
-
-        return Ok(Some(ThreadContext {
-            info: thread_info,
-            first_message: ThreadContextFirstMessage {
-                is_mentioning_bot,
+        return Ok(Some(InteractionContext {
+            thread_info,
+            trigger: InteractionTrigger {
+                is_mentioning_bot: current_event_is_mentioning_bot,
                payload: current_event_payload.clone(),
            },
        }));
    };

-    let Relation::Thread(thread) = relation else {
-        // This is a reply or a replacement, etc. It's not a thread.
-        // We don't care about this.
-        return Ok(None);
-    };
+    match relation {
+        Relation::Thread(thread) => {
+            determine_interaction_context_for_room_event_related_to_thread(
+                bot_user_id,
+                bot_display_name,
+                room,
+                current_event,
+                event_fetcher,
+                current_event_is_mentioning_bot,
+                thread,
+            )
+            .await
+        }
+        Relation::Reply { in_reply_to } => {
+            determine_interaction_context_for_room_event_related_to_reply(
+                current_event,
+                current_event_is_mentioning_bot,
+                in_reply_to.event_id.clone(),
+            )
+            .await
+        }

+        // This is a replacement or something else. It's not something we support.
+        _ => return Ok(None),
+    }
+}
+
+async fn determine_interaction_context_for_room_event_related_to_thread(
+    bot_user_id: &OwnedUserId,
+    bot_display_name: &Option<String>,
+    room: &Room,
+    current_event: &OriginalSyncRoomMessageEvent,
+    event_fetcher: &Arc<RoomEventFetcher>,
+    current_event_is_mentioning_bot: bool,
+    thread: &Thread,
+) -> anyhow::Result<Option<InteractionContext>> {
    let thread_info = ThreadInfo::new(thread.event_id.clone(), current_event.event_id.clone());

+    tracing::trace!(
+        ?current_event_is_mentioning_bot,
+        is_thread_root_only = thread_info.is_thread_root_only(),
+        "Dealing with a thread reply",
+    );
+
+    if current_event_is_mentioning_bot && !thread_info.is_thread_root_only() {
+        // If the current event is a thread reply and is mentioning the bot,
+        // it's probably someone trying to involve us in the threaded conversation.
+        // See: https://github.com/etkecc/baibot/issues/15
+        //
+        // In such cases, we don't care what the thread root event is like or what the current event is like,
+        // we want text-generation to be triggered for this whole thread regardless.
+        return Ok(Some(InteractionContext {
+            thread_info,
+            trigger: InteractionTrigger {
+                is_mentioning_bot: true,
+                payload: MessagePayload::SynthethicChatCompletionTriggerInThread,
+            },
+        }));
+    }
+
    let start_time = std::time::Instant::now();

    let thread_start_timeline_event = event_fetcher
@@ -183,82 +387,47 @@ pub async fn determine_thread_context_for_room_event(
        "Fetched thread start event"
    );

-    let thread_start_timeline_event_deserialized =
-        match thread_start_timeline_event.event.deserialize() {
-            Ok(value) => value,
-            Err(err) => {
-                return Err(anyhow::format_err!(
-                    "Failed to deserialize thread start event {}: {:?}",
-                    thread.event_id,
-                    err
-                ));
-            }
-        };
+    let thread_start_detailed_message_payload = timeline_event_to_detailed_message_payload(
+        &thread.event_id,
+        thread_start_timeline_event,
+        thread_info.clone(),
+        bot_user_id,
+        bot_display_name,
+    )?;

-    let AnyTimelineEvent::MessageLike(thread_start_message_like_event) =
-        thread_start_timeline_event_deserialized
-    else {
-        tracing::trace!(
-            "Ignoring non-MessageLike thread start event: {:?}",
-            thread_start_timeline_event_deserialized
-        );
+    let Some(detailed_message_payload) = thread_start_detailed_message_payload else {
        return Ok(None);
    };

-    let (thread_start_message_is_mentioning_bot, thread_start_message_payload) =
-        match thread_start_message_like_event {
-            AnyMessageLikeEvent::RoomEncrypted(room_message) => {
-                tracing::warn!(
-                    "Could not inspect thread start event {} because it failed to decrypt: {:?}",
-                    thread.event_id.clone(),
-                    room_message
-                );
+    Ok(Some(InteractionContext {
+        thread_info,
+        trigger: InteractionTrigger {
+            is_mentioning_bot: detailed_message_payload.is_mentioning_bot,
+            payload: detailed_message_payload.message_payload,
+        },
+    }))
+}

-                // There's no way to know and it doesn't matter anyway.
-                let is_mentioning_bot = false;
+async fn determine_interaction_context_for_room_event_related_to_reply(
+    current_event: &OriginalSyncRoomMessageEvent,
+    current_event_is_mentioning_bot: bool,
+    reply_to_event_id: OwnedEventId,
+) -> anyhow::Result<Option<InteractionContext>> {
+    tracing::trace!(?current_event_is_mentioning_bot, "Dealing with a reply");

-                (
-                    is_mentioning_bot,
-                    MessagePayload::Encrypted(thread_info.clone()),
-                )
-            }
-            AnyMessageLikeEvent::RoomMessage(room_message) => {
-                if let MessageLikeEvent::Original(room_message_original) = room_message {
-                    let room_message_payload: Result<MessagePayload, String> =
-                        room_message_original.content.msgtype.clone().try_into();
+    if !current_event_is_mentioning_bot {
+        // If the current event is not mentioning the bot, we don't care about it.
+        tracing::trace!("Ignoring reply event which does not mention the bot");
+        return Ok(None);
+    }

-                    let Ok(room_message_payload) = room_message_payload else {
-                        tracing::debug!(
-                            msg_type = room_message_original.content.msgtype(),
-                            "Ignoring thread start message of unknown type",
-                        );
-                        return Ok(None);
-                    };
+    let thread_info = ThreadInfo::new(reply_to_event_id.clone(), current_event.event_id.clone());

-                    let is_mentioning_bot =
-                        is_event_mentioning_bot(&room_message_original.content, bot_user_id);
-
-                    (is_mentioning_bot, room_message_payload)
-                } else {
-                    tracing::error!("Ignoring thread start message which appears to be redacted");
-
-                    return Ok(None);
-                }
-            }
-            other => {
-                tracing::trace!(
-                    "Ignoring unknown MessageLike thread start event: {:?}",
-                    other
-                );
-                return Ok(None);
-            }
-        };
-
-    Ok(Some(ThreadContext {
-        info: thread_info,
-        first_message: ThreadContextFirstMessage {
-            is_mentioning_bot: thread_start_message_is_mentioning_bot,
-            payload: thread_start_message_payload,
+    Ok(Some(InteractionContext {
+        thread_info,
+        trigger: InteractionTrigger {
+            is_mentioning_bot: true,
+            payload: MessagePayload::SynthethicChatCompletionTriggerForReply,
        },
    }))
 }
@@ -266,6 +435,7 @@ pub async fn determine_thread_context_for_room_event(
 fn is_event_mentioning_bot(
    event_content: &RoomMessageEventContent,
    bot_user_id: &OwnedUserId,
+    bot_display_name: &Option<String>,
 ) -> bool {
    if let Some(mentions) = &event_content.mentions {
        mentions
@@ -277,11 +447,138 @@ fn is_event_mentioning_bot(
        // (see https://spec.matrix.org/latest/client-server-api/#user-and-room-mentions),
        // we also do string matching here.
        //
-        // It may be even better to match not only against the MXID, but also against the bot's
-        // room-specific display name.
+        // As of 2024-10-03, at least Element iOS does not support the new Mentions specification
+        // and is still quite widespread.
        //
        // We may consider dropping this string-matching behavior altogether in the future,
        // so improving this compatibility block is not a high priority.
-        event_content.body().contains(bot_user_id.as_str())
+        if event_content.body().contains(bot_user_id.as_str()) {
+            return true;
+        }
+
+        if let Some(bot_display_name) = bot_display_name {
+            return event_content.body().contains(bot_display_name);
+        }
+
+        false
    }
 }
+
+fn timeline_event_to_detailed_message_payload(
+    timeline_event_id: &OwnedEventId,
+    timeline_event: TimelineEvent,
+    thread_info: ThreadInfo,
+    bot_user_id: &OwnedUserId,
+    bot_display_name: &Option<String>,
+) -> anyhow::Result<Option<DetailedMessagePayload>> {
+    let timeline_event_deserialized = match timeline_event.event.deserialize() {
+        Ok(value) => value,
+        Err(err) => {
+            return Err(anyhow::format_err!(
+                "Failed to deserialize timeline event {}: {:?}",
+                timeline_event_id,
+                err
+            ));
+        }
+    };
+
+    let AnyTimelineEvent::MessageLike(thread_start_message_like_event) =
+        timeline_event_deserialized
+    else {
+        tracing::trace!(
+            "Ignoring non-MessageLike timeline event: {:?}",
+            timeline_event_deserialized
+        );
+        return Ok(None);
+    };
+
+    let (is_mentioning_bot, message_payload) = match thread_start_message_like_event {
+        AnyMessageLikeEvent::RoomEncrypted(room_message) => {
+            tracing::warn!(
+                "Could not inspect event {} because it failed to decrypt: {:?}",
+                timeline_event_id.clone(),
+                room_message
+            );
+
+            // There's no way to know and it doesn't matter anyway.
+            let is_mentioning_bot = false;
+
+            (
+                is_mentioning_bot,
+                MessagePayload::Encrypted(thread_info.clone()),
+            )
+        }
+        AnyMessageLikeEvent::RoomMessage(room_message) => {
+            if let MessageLikeEvent::Original(room_message_original) = room_message {
+                let room_message_payload: Result<MessagePayload, String> =
+                    room_message_original.content.msgtype.clone().try_into();
+
+                let Ok(room_message_payload) = room_message_payload else {
+                    tracing::debug!(
+                        msg_type = room_message_original.content.msgtype(),
+                        "Ignoring event message of unknown type",
+                    );
+                    return Ok(None);
+                };
+
+                let is_mentioning_bot = is_event_mentioning_bot(
+                    &room_message_original.content,
+                    bot_user_id,
+                    bot_display_name,
+                );
+
+                (is_mentioning_bot, room_message_payload)
+            } else {
+                tracing::error!("Ignoring event message which appears to be redacted");
+
+                return Ok(None);
+            }
+        }
+        other => {
+            tracing::trace!("Ignoring unknown MessageLike event: {:?}", other);
+            return Ok(None);
+        }
+    };
+
+    Ok(Some(DetailedMessagePayload {
+        is_mentioning_bot,
+        message_payload,
+    }))
+}
+
+/// Creates a list of prefixes to strip from the beginning of message texts that mention the bot user.
+///
+/// Different clients do mentions differently.
+/// The body text containing the mention usually contains one of:
+/// - the full user ID (includes a @ prefix by default)
+/// - the localpart (with a @ prefix)
+/// - the localpart (without a @ prefix)
+/// - the display name (with a @ prefix)
+/// - the display name (without a @ prefix)
+///
+/// Some add a `: ` suffix after the mention.
+///
+/// There's no guarantee that the mention is at the start even.
+/// It being there is most common and we try to strip it from there
+/// as best as we can.
+pub fn create_list_of_bot_user_prefixes_to_strip(
+    bot_user_id: &OwnedUserId,
+    bot_display_name: &Option<String>,
+) -> Vec<String> {
+    let bot_user_id_localpart = bot_user_id.localpart();
+
+    let mut prefixes_to_strip = vec![
+        bot_user_id.as_str().to_owned(),
+        format!("@{}", bot_user_id_localpart),
+        bot_user_id_localpart.to_owned(),
+    ];
+
+    if let Some(bot_display_name) = bot_display_name {
+        prefixes_to_strip.push(format!("@{}", bot_display_name));
+        prefixes_to_strip.push(bot_display_name.to_owned());
+    }
+
+    prefixes_to_strip.push(":".to_owned());
+
+    prefixes_to_strip
+}
--- a/src/conversation/matrix/utils/tests.rs
+++ b/src/conversation/matrix/utils/tests.rs
@@ -1,29 +1,42 @@
+use chrono::{TimeZone, Utc};
+
+use mxlink::matrix_sdk::ruma::OwnedUserId;
+
 use crate::conversation::matrix::{
    MatrixMessage, MatrixMessageProcessingParams, MatrixMessageType,
 };

 #[test]
 fn is_message_from_allowed_sender() {
-    let bot_user_id = "@bot:example.com";
-    let allowed_user_id = "@user.someone:example.com";
-    let unallowed_user_id = "@another:example.com";
+    let bot_user_id =
+        OwnedUserId::try_from("@bot:example.com").expect("Failed to parse bot user ID");
+    let allowed_user_id = OwnedUserId::try_from("@user.someone:example.com").unwrap();
+    let unallowed_user_id = OwnedUserId::try_from("@another:example.com").unwrap();
+
+    let timestamp = Utc.with_ymd_and_hms(2024, 9, 20, 18, 34, 15).unwrap();

    let bot_message = MatrixMessage {
        sender_id: bot_user_id.to_owned(),
        message_type: MatrixMessageType::Text,
        message_text: "Hello!".to_owned(),
+        mentioned_users: vec![],
+        timestamp,
    };

    let allowed_user_message = MatrixMessage {
        sender_id: allowed_user_id.to_owned(),
        message_type: MatrixMessageType::Text,
        message_text: "Hello!".to_owned(),
+        mentioned_users: vec![],
+        timestamp,
    };

    let unallowed_user_message = MatrixMessage {
        sender_id: unallowed_user_id.to_owned(),
        message_type: MatrixMessageType::Text,
        message_text: "Hello!".to_owned(),
+        mentioned_users: vec![],
+        timestamp,
    };

    let parsed_regex = match mxidwc::parse_pattern("@user.*:example.com") {
@@ -36,65 +49,106 @@ fn is_message_from_allowed_sender() {
    let allowed_users = vec![parsed_regex];

    assert!(
-        super::is_message_from_allowed_sender(&bot_message, bot_user_id, &vec![]),
+        super::is_message_from_allowed_sender(&bot_message, &bot_user_id, Some(&allowed_users)),
        "Bot message should be allowed"
    );

    assert!(
-        super::is_message_from_allowed_sender(&allowed_user_message, bot_user_id, &allowed_users),
+        super::is_message_from_allowed_sender(
+            &allowed_user_message,
+            &bot_user_id,
+            Some(&allowed_users)
+        ),
        "Allowed user message should be allowed"
    );

    assert!(
        !super::is_message_from_allowed_sender(
            &unallowed_user_message,
-            bot_user_id,
-            &allowed_users
+            &bot_user_id,
+            Some(&allowed_users),
        ),
        "Unallowed user message should be ignored"
    );
+
+    assert!(
+        super::is_message_from_allowed_sender(&unallowed_user_message, &bot_user_id, None,),
+        "An empty list of allowed users lets everyone through"
+    );
 }

 #[tokio::test]
-async fn process_matrix_messages_in_thread() {
-    let bot_user_id = "@bot:example.com";
-    let allowed_user_id = "@user.someone:example.com";
-    let unallowed_user_id = "@another:example.com";
+async fn process_matrix_messages() {
+    let bot_user_id =
+        OwnedUserId::try_from("@bot:example.com").expect("Failed to parse bot user ID");
+    let allowed_user_id = OwnedUserId::try_from("@user.someone:example.com").unwrap();
+    let unallowed_user_id = OwnedUserId::try_from("@another:example.com").unwrap();
+
+    let timestamp = Utc.with_ymd_and_hms(2024, 9, 20, 18, 34, 15).unwrap();

    let allowed_user_message = MatrixMessage {
        sender_id: allowed_user_id.to_owned(),
        message_type: MatrixMessageType::Text,
        message_text: "Hello from the user!".to_owned(),
+        mentioned_users: vec![],
+        timestamp,
    };

    let allowed_user_message_with_prefix = MatrixMessage {
        sender_id: allowed_user_id.to_owned(),
        message_type: MatrixMessageType::Text,
        message_text: "!bai Hello from the user!".to_owned(),
+        mentioned_users: vec![],
+        timestamp,
    };

    let allowed_user_message_with_prefix_no_space = MatrixMessage {
        sender_id: allowed_user_id.to_owned(),
        message_type: MatrixMessageType::Text,
        message_text: "!baiHello from the user!".to_owned(),
+        mentioned_users: vec![],
+        timestamp,
    };

    let allowed_user_message_with_prefix_full_width_space = MatrixMessage {
        sender_id: allowed_user_id.to_owned(),
        message_type: MatrixMessageType::Text,
        message_text: "!bai　Hello from the user!".to_owned(),
+        mentioned_users: vec![],
+        timestamp,
    };

    let bot_message = MatrixMessage {
        sender_id: bot_user_id.to_owned(),
        message_type: MatrixMessageType::Text,
        message_text: "Hello from the bot!".to_owned(),
+        mentioned_users: vec![],
+        timestamp,
+    };
+
+    let allowed_user_message_with_bot_mention = MatrixMessage {
+        sender_id: allowed_user_id.to_owned(),
+        message_type: MatrixMessageType::Text,
+        message_text: "@baibot: Hello from the user!".to_owned(),
+        mentioned_users: vec![bot_user_id.to_owned()],
+        timestamp,
+    };
+
+    // The message text is the same as above - it mentions the bot, but the actually-mentioned user is another user.
+    let allowed_user_message_with_another_user_mention = MatrixMessage {
+        sender_id: allowed_user_id.to_owned(),
+        message_type: MatrixMessageType::Text,
+        message_text: allowed_user_message_with_bot_mention.message_text.clone(),
+        mentioned_users: vec![allowed_user_id.to_owned()],
+        timestamp,
    };

    let unallowed_user_message = MatrixMessage {
        sender_id: unallowed_user_id.to_owned(),
        message_type: MatrixMessageType::Text,
        message_text: "Hello from an unallowed user!".to_owned(),
+        mentioned_users: vec![],
+        timestamp,
    };

    let parsed_regex = match mxidwc::parse_pattern("@user.*:example.com") {
@@ -106,12 +160,24 @@ async fn process_matrix_messages_in_thread() {

    let allowed_users = vec![parsed_regex];

-    let message_processing_params_basic =
-        super::MatrixMessageProcessingParams::new(bot_user_id.to_owned(), allowed_users.clone());
+    let message_processing_params_basic = super::MatrixMessageProcessingParams::new(
+        bot_user_id.to_owned(),
+        Some(allowed_users.clone()),
+    );

    let message_processing_params_with_prefix_stripping =
-        super::MatrixMessageProcessingParams::new(bot_user_id.to_owned(), allowed_users.clone())
-            .with_first_message_stripped_prefixes(vec!["!bai".to_owned()]);
+        super::MatrixMessageProcessingParams::new(
+            bot_user_id.to_owned(),
+            Some(allowed_users.clone()),
+        )
+        .with_first_message_prefixes_to_strip(vec!["!bai".to_owned()]);
+
+    let message_processing_params_with_bot_user_prefix_stripping =
+        super::MatrixMessageProcessingParams::new(
+            bot_user_id.to_owned(),
+            Some(allowed_users.clone()),
+        )
+        .with_bot_user_prefixes_to_strip(vec!["@baibot: ".to_owned(), "@baibot".to_owned()]);

    struct TestCase {
        name: String,
@@ -195,10 +261,23 @@ async fn process_matrix_messages_in_thread() {
                "!bai Hello from the user!".to_owned(),
            ],
        },
+        TestCase {
+            name: "Messages that mention the bot user get the bot user prefix stripped"
+                .to_owned(),
+            messages: vec![
+                allowed_user_message_with_bot_mention.clone(),
+                allowed_user_message_with_another_user_mention.clone(),
+            ],
+            message_processing_params: message_processing_params_with_bot_user_prefix_stripping.clone(),
+            expected_message_texts: vec![
+                "Hello from the user!".to_owned(),
+                "@baibot: Hello from the user!".to_owned(),
+            ],
+        },
    ];

    for test_case in test_cases {
-        let processed_messages = super::process_matrix_messages_in_thread(
+        let processed_messages = super::process_matrix_messages(
            &test_case.messages,
            &test_case.message_processing_params,
        )
@@ -216,3 +295,41 @@ async fn process_matrix_messages_in_thread() {
        );
    }
 }
+
+#[test]
+fn create_list_of_bot_user_prefixes_to_strip() {
+    let bot_user_id =
+        OwnedUserId::try_from("@baibot:example.com").expect("Failed to parse bot user ID");
+
+    // Test case 1: Bot user with no display name
+    let bot_display_name = None;
+    let prefixes =
+        super::create_list_of_bot_user_prefixes_to_strip(&bot_user_id, &bot_display_name);
+
+    assert_eq!(
+        prefixes,
+        vec![
+            "@baibot:example.com".to_string(),
+            "@baibot".to_string(),
+            "baibot".to_string(),
+            ":".to_string()
+        ]
+    );
+
+    // Test case 2: Bot user with display name
+    let bot_display_name = Some("Assistant".to_string());
+    let prefixes =
+        super::create_list_of_bot_user_prefixes_to_strip(&bot_user_id, &bot_display_name);
+
+    assert_eq!(
+        prefixes,
+        vec![
+            "@baibot:example.com".to_string(),
+            "@baibot".to_string(),
+            "baibot".to_string(),
+            "@Assistant".to_string(),
+            "Assistant".to_string(),
+            ":".to_string()
+        ]
+    );
+}
--- a/src/conversation/matrix_llm_bridge.rs
+++ b/src/conversation/matrix_llm_bridge.rs
@@ -1,9 +1,14 @@
+use std::sync::Arc;
+
 use mxlink::matrix_sdk::ruma::OwnedEventId;
 use mxlink::MatrixLink;

+use crate::conversation::matrix::MatrixMessage;
+
 use super::llm::{convert_matrix_message_to_llm_message, Conversation, Message};
 use super::matrix::{
-    get_matrix_messages_in_thread, process_matrix_messages_in_thread, MatrixMessageProcessingParams,
+    get_matrix_messages_in_reply_chain, get_matrix_messages_in_thread, process_matrix_messages,
+    MatrixMessageProcessingParams, RoomEventFetcher,
 };

 pub async fn create_llm_conversation_for_matrix_thread(
@@ -14,7 +19,33 @@ pub async fn create_llm_conversation_for_matrix_thread(
 ) -> Result<Conversation, mxlink::matrix_sdk::Error> {
    let messages = get_matrix_messages_in_thread(matrix_link, room, thread_id).await?;

-    let messages_filtered = process_matrix_messages_in_thread(&messages, params).await;
+    let llm_messages = filter_messages_and_convert_to_llm_messages(messages, params).await;
+
+    Ok(Conversation {
+        messages: llm_messages,
+    })
+}
+
+pub async fn create_llm_conversation_for_matrix_reply_chain(
+    event_fetcher: &Arc<RoomEventFetcher>,
+    room: &mxlink::matrix_sdk::Room,
+    event_id: OwnedEventId,
+    params: &MatrixMessageProcessingParams,
+) -> Result<Conversation, mxlink::matrix_sdk::Error> {
+    let messages = get_matrix_messages_in_reply_chain(event_fetcher, room, event_id).await?;
+
+    let llm_messages = filter_messages_and_convert_to_llm_messages(messages, params).await;
+
+    Ok(Conversation {
+        messages: llm_messages,
+    })
+}
+
+async fn filter_messages_and_convert_to_llm_messages(
+    messages: Vec<MatrixMessage>,
+    params: &MatrixMessageProcessingParams,
+) -> Vec<Message> {
+    let messages_filtered = process_matrix_messages(&messages, params).await;

    let mut llm_messages: Vec<Message> = Vec::new();

@@ -28,7 +59,5 @@ pub async fn create_llm_conversation_for_matrix_thread(
        llm_messages.push(llm_message);
    }

-    Ok(Conversation {
-        messages: llm_messages,
-    })
+    llm_messages
 }
--- a/src/conversation/mod.rs
+++ b/src/conversation/mod.rs
@@ -2,4 +2,6 @@ pub(crate) mod llm;
 pub(crate) mod matrix;
 mod matrix_llm_bridge;

-pub(crate) use matrix_llm_bridge::create_llm_conversation_for_matrix_thread;
+pub(crate) use matrix_llm_bridge::{
+    create_llm_conversation_for_matrix_reply_chain, create_llm_conversation_for_matrix_thread,
+};
--- a/src/entity/interaction_context.rs
+++ b/src/entity/interaction_context.rs
@@ -0,0 +1,13 @@
+use mxlink::ThreadInfo;
+
+use super::MessagePayload;
+
+pub struct InteractionContext {
+    pub thread_info: ThreadInfo,
+    pub trigger: InteractionTrigger,
+}
+
+pub struct InteractionTrigger {
+    pub is_mentioning_bot: bool,
+    pub payload: MessagePayload,
+}
--- a/src/entity/message_context.rs
+++ b/src/entity/message_context.rs
@@ -15,6 +15,8 @@ pub struct MessageContext {
    admin_whitelist_regexes: Vec<regex::Regex>,
    trigger_event_info: TriggerEventInfo,
    thread_info: ThreadInfo,
+
+    bot_display_name: Option<String>,
 }

 impl MessageContext {
@@ -31,9 +33,20 @@ impl MessageContext {
            admin_whitelist_regexes,
            trigger_event_info,
            thread_info,
+
+            bot_display_name: None,
        }
    }

+    pub fn with_bot_display_name(mut self, value: Option<String>) -> Self {
+        self.bot_display_name = value;
+        self
+    }
+
+    pub fn bot_display_name(&self) -> &Option<String> {
+        &self.bot_display_name
+    }
+
    pub fn room(&self) -> &Room {
        &self.room
    }
--- a/src/entity/message_payload.rs
+++ b/src/entity/message_payload.rs
@@ -6,10 +6,28 @@ use mxlink::matrix_sdk::ruma::{OwnedEventId, OwnedUserId};
 use mxlink::ThreadInfo;

 /// MessagePayload is like matrix-sdk's MessageType, but represents only message types that the bot deals with and payloads are massaged a bit.
+///
+/// This also includes a few synthetic events.
 #[derive(Debug, Clone)]
 pub enum MessagePayload {
-    Text(TextMessageEventContent),
+    /// A synthetic message payload that indicates that the bot should produce a reply inside a thread.
+    /// This does not represent an actual message event, it's just a way to trigger a chat completion.
+    ///
+    /// When this is invoked, the ThreadInfo contains the full thread details (which represents our context).
+    ///
+    /// See: https://github.com/etkecc/baibot/issues/15
+    SynthethicChatCompletionTriggerInThread,

+    /// A synthetic message payload that indicates that the bot should produce a reply to a specific message.
+    /// This does not represent an actual message event, it's just a way to trigger a chat completion.
+    ///
+    /// When this is invoked, the ThreadInfo would refer to the reply-message that triggered us.
+    /// We can follow the chain upward from it to get the full context.
+    ///
+    /// See: https://github.com/etkecc/baibot/issues/15
+    SynthethicChatCompletionTriggerForReply,
+
+    Text(TextMessageEventContent),
    Audio(AudioMessageEventContent),

    Reaction {
--- a/src/entity/mod.rs
+++ b/src/entity/mod.rs
@@ -1,15 +1,15 @@
 pub mod catch_up_marker;
 pub mod cfg;
 pub mod globalconfig;
+mod interaction_context;
 mod message_context;
 mod message_payload;
 mod room_config_context;
 pub mod roomconfig;
-mod thread_context;
 mod trigger_event_info;

+pub use interaction_context::{InteractionContext, InteractionTrigger};
 pub use message_context::MessageContext;
 pub use message_payload::MessagePayload;
 pub use room_config_context::RoomConfigContext;
-pub use thread_context::{ThreadContext, ThreadContextFirstMessage};
 pub use trigger_event_info::TriggerEventInfo;
--- a/src/entity/thread_context.rs
+++ b/src/entity/thread_context.rs
@@ -1,13 +0,0 @@
-use mxlink::ThreadInfo;
-
-use super::MessagePayload;
-
-pub struct ThreadContext {
-    pub info: ThreadInfo,
-    pub first_message: ThreadContextFirstMessage,
-}
-
-pub struct ThreadContextFirstMessage {
-    pub is_mentioning_bot: bool,
-    pub payload: MessagePayload,
-}
--- a/src/utils/text_to_speech.rs
+++ b/src/utils/text_to_speech.rs
@@ -5,12 +5,18 @@ use super::text::{block_quote, block_unquote};
 /// Creates a text message which is based on transcribed audio.
 /// This text message is prefixed with an emoji and blockquoted, to indicate that it is a transcription.
 /// To reverse the process, use `parse_transcribed_message_text()`.
+///
+/// It should be noted that in certain cases (Transcribe-only mode), transcriptions are posted as regular notice messages which do not include
+/// the `> 🦻` prefixing. That is, not every transcribed message will pass through here (intentionally).
 pub fn create_transcribed_message_text(text: &str) -> String {
    block_quote(&format!("{} {}", AgentPurpose::SpeechToText.emoji(), text))
 }

 /// Parses a transcribed message text, reversing the process done by `create_transcribed_message_text()`.
 /// If the provided text string does not match the expected format, None is returned.
+///
+/// It should be noted that in certain cases (Transcribe-only mode), transcriptions are posted as regular notice messages which do not include
+/// the `> 🦻` prefixing. This function will not handle these properly.
 pub fn parse_transcribed_message_text(text: &str) -> Option<String> {
    if !text.starts_with("> ") {
        return None;
Author	SHA1	Message	Date
Slavi Pantaleev	4a5966401c	Release 1.3.2	2024-11-12 10:37:49 +02:00
Slavi Pantaleev	d92dfba2bf	Upgrade Rust compiler in container image (1.81.0 -> 1.82.0)	2024-11-12 10:37:16 +02:00
Slavi Pantaleev	8538d6b2b8	Upgrade component services	2024-11-12 10:36:54 +02:00
Slavi Pantaleev	23f763ba72	Update dependencies	2024-11-12 10:22:26 +02:00
Slavi Pantaleev	a9e4ab1bdb	Release 1.3.1	2024-10-03 16:30:58 +03:00
Slavi Pantaleev	d9a045a5e4	Make fallback user mentions support also match against the bot's room-specific username It seems like Element iOS benefits from this.	2024-10-03 16:28:49 +03:00
Slavi Pantaleev	393be9be5a	Remove strip_rich_reply_fallback_text in favor of remove_plain_reply_fallback from ruma events No need to reinvent the wheel.	2024-10-03 16:02:06 +03:00
Slavi Pantaleev	a7b016a3d3	Release 1.3.0	2024-10-03 12:08:00 +03:00
Slavi Pantaleev	85e66406dc	Allow for prompt caching to work by using `baibot_conversation_start_time_utc` instead of `baibot_now_utc` This patch introduces a new `baibot_conversation_start_time_utc` variable which indicates the time the conversation got started. Using `baibot_now_utc` is still possible, but given that the current time is a moving target, its use is in conflict with prompt caching. Because the new `baibot_conversation_start_time_utc` prompt variable is a more reasonable default, we're now using it in all sample configs.	2024-10-03 11:48:14 +03:00
Slavi Pantaleev	db9422740c	Add support for OpenAI's o1 models by making `max_response_tokens` optional The other prerequisite seems to be not using a `prompt` (`prompt: null`), but we already supported this. It'd be nice to add an optional `max_completion_tokens` parameter as well, for the benefit of the o1 models, but this is not yet supported by async-openai. Possibly tracked here: https://github.com/64bit/async-openai/issues/272	2024-10-03 10:36:28 +03:00
Slavi Pantaleev	90fbad5b64	Update sample & default OpenAI provider configs to use gpt-4o (instead of gpt-4o-2024-08-06) Since 2024-10-02, `gpt-4o` is actually the same as `gpt-4o-2024-08-06`. We previously used `gpt-4o-2024-08-06`, because it was pointing to a much better (longer context) model. Since they're both the same now, we'd better stick to the unpinned model and make it easier for future users to get upgrades.	2024-10-03 09:26:41 +03:00
Slavi Pantaleev	b40226826f	Restore fallback support for user mentions Fallback support was intentionally removed in `9908512968`, because it was deemed OK to do so. It turns out that Element iOS still doesn't properly do user mentions (and likely never will, until Element X replaces it), so we can't just drop the fallback user mentions logic without affecting all these clients. It's possible that the Element Android is no better (unverified claim).	2024-10-03 09:18:02 +03:00
Slavi Pantaleev	b89f0db71a	Relocate "On-demand involvement" feature description section [skip ci]	2024-10-02 09:12:12 +03:00
Slavi Pantaleev	36fdb46633	Release 1.2.0	2024-10-01 22:01:17 +03:00
Slavi Pantaleev	04ce8db1fc	Update dependencies	2024-10-01 21:35:49 +03:00
Slavi Pantaleev	9908512968	Add support for on-demand involvement Fixes https://github.com/etkecc/baibot/issues/15	2024-10-01 21:06:54 +03:00
Slavi Pantaleev	eae6472c7a	Upgrade services	2024-10-01 18:33:08 +03:00
Slavi Pantaleev	e6aa956423	Do not send blockquote-formatted transcription when replying without a thread This actually fixes 2 issues. Fixes https://github.com/etkecc/baibot/issues/14 Fixes https://github.com/etkecc/baibot/issues/17 When people enable transcribe-only mode and the bot replies outside of a thread, messages will no longer look like this: `> 🦻 Transcribed text`. Instead, they will: - look like this: `Transcribed text` - get an emoji reaction (🦻) sent by the bot itself, to indicate that the message is a transcription --------------------------------------- As https://github.com/etkecc/baibot/issues/14 discusses, the `> 🦻` prefixing of messages also served the purpose of indicating to the bot that this is not its own message, but rather something it "heard" from a user. Given that out-of-thread replies no longer include this, they could be mistaken for bot messages. Because transcribed messages are posted as notice messages, we can easily tell them apart from regular text-generated messages by the bot itself, so we can (and do) treat them differently. Thankfully, the bot does not yet support building a text-generation conversation from arbitrary messages (something discussed in https://github.com/etkecc/baibot/issues/15), so these out-of-thread replies having the wrong owner are not an issue for now. If we do land support for this, we'll probably need to make the bot inspect such notice messages posted by it, inspect their reactons and attribute them properly (🦻 -> user message).	2024-09-30 17:35:14 +03:00
Slavi Pantaleev	7a38216192	Upgrade services	2024-09-26 22:32:34 +03:00
Slavi Pantaleev	d522d268e2	Use the same variable-infused prompt for the ollama agent in etc/app/config.yml.dist	2024-09-26 22:32:28 +03:00
Slavi Pantaleev	72120c5dc2	Explicitly keep authenticated media disabled in the Synapse configuration [skip ci] Related to https://github.com/etkecc/baibot/issues/12	2024-09-23 06:09:32 +00:00
Slavi Pantaleev	a2c35238c2	Upgrade services [skip ci]	2024-09-23 06:08:17 +00:00
Slavi Pantaleev	97f5cbb00b	Fix example for `baibot_now_utc` prompt variable This feature went through a few iterations. At some point, a `(local timezone/time: unknown)` suffix was part of the `baibot_now_utc` variable (hoping it improves the model's awareness that it doesn't know the current local time), but the suffix was ultimately removed as unnecessary.	2024-09-22 08:52:19 +00:00