docs: auto-sync with merged PRs (through 2026-09-12)

packages/kilo-docs/pages/code-with-ai/agents/context-mentions.md

@@ -22,6 +22,7 @@ Type `@` in the chat input to get autocomplete suggestions. You can mention:
 
 | Mention | Description | Example |
 |---|---|---|
+| **Model** | Insert an inline `@provider/model` reference for Agent Manager sessions and subagents | `@` → **Model** → pick a model |
 | **File** | Attach a file's contents to your message | `@src/utils.ts` |
 | **Terminal** | In the sidebar, include the active VS Code terminal output. In Agent Manager, include the focused terminal for the selected session or worktree, including embedded **Run** and **Setup** tabs. | `@terminal` |
 | **Git Changes** | Attach uncommitted working-tree diffs and new files | `@git-changes` |
@@ -31,6 +32,10 @@ Selecting a suggestion inserts the mention and highlights it in the input. File
 
 Terminal context is limited to 500 lines or 50,000 characters. Longer output is truncated. If no terminal content is available, Kilo reports that terminal content is unavailable instead of attaching an empty file.
 
+### Referencing a model
+
+Choose **Model** from the `@` menu to insert a model reference such as `@anthropic/claude-sonnet-4`. Use the reference to ask Kilo to run an Agent Manager session or subagent with that model. The reference does not attach a file or change the current session's model.
+
 ### Referencing Past Chats
 
 Choosing **Past chats** from the `@` menu opens a searchable picker of sessions in the current workspace and related Git worktrees, including Agent Manager worktrees, ordered by recency. Search matches session titles and worktree names. Selecting a session inserts a highlighted mention token; when you send the message, that session's current chat history is added as context so the agent can build on the earlier conversation. Clicking the mention token opens the referenced session.

packages/kilo-docs/pages/code-with-ai/agents/model-selection.md

@@ -11,17 +11,19 @@ Instead of maintaining a static list that's perpetually behind, we built somethi
 
 ## Model Routing and Configuration
 
-Kilo's IDE Extension and CLI configurations have four separate, independently configurable model slots:
+You can configure models separately for different tasks in the VS Code extension and kilo CLI:
 
 - **Main model** — the primary model your agent uses for coding tasks, chat, and reasoning. This is what you pick with the model selector, `/models`, or the `model` key in `kilo.jsonc`. See [How to Select and Switch Models](#how-to-select-and-switch-models) for the full precedence order and per-agent config.
-  - The main model is also used for context compaction/summarization and todo-list generation.
+  - The main model is also used for todo-list generation, and for context compaction/summarization when no compaction model is set.
 - **Small model** — a lightweight model used for session title generation, commit message generation, and prompt enhancement. Configured with the `small_model` key in `kilo.jsonc`, or the **Small Model** field on the **Settings → Models** tab.
   - If left unset, Kilo resolves it according to the following logic: 
     1. Find a small/cheap variant on your current provider (e.g. Haiku on Anthropic, Flash on Gemini).
     2. Fall back to [`kilo-auto/small`](/docs/gateway/models-and-providers#kilo-autosmall) if the Kilo Gateway provider is authenticated in your session.
     3. Reuse your main model if you're not authenticated to the Kilo Gateway.
 - **Subagent model** — the default model for subagents launched by the `task` tool. Configured with the `subagent_model` key in `kilo.jsonc`, or the **Subagent Model** field on the **Settings → Models** tab.
   - If left unset, inherits whichever model the parent agent session is currently using.
+- **Compaction model** — the model used for context compaction/summarization. Configured with the `agent.compaction.model` key in `kilo.jsonc`, or the **Compaction model** field on the **Settings → Models** tab.
+  - If left unset, the current session's model is used. See [Use a different model for compaction](/docs/customize/context/context-condensing#use-a-different-model-for-compaction).
 - **Autocomplete model** — the model used for inline code completions as you type. See [Autocomplete: Provider and Model](/docs/code-with-ai/features/autocomplete#provider-and-model) for how to configure it.
 
 ### Configuring Local Usage

packages/kilo-docs/pages/code-with-ai/platforms/mobile.md

@@ -62,6 +62,15 @@ The composer stays editable while the agent is working, so you don't have to wai
 
 A queued message shows a subtle **Queued** badge on its bubble. The badge clears when the message starts processing or when the queue drains or is cancelled. Queueing works for Cloud Agent sessions and for remote sessions on a connected `kilo remote` CLI instance.
 
+## Voice input
+
+Every voice dictation in the app uses one of two transcription engines:
+
+- **Operating system** (default) — your device's built-in speech recognizer.
+- **Kilo gateway** — transcribes through your Kilo account with a transcription model you choose. Enable the **Gateway transcription** switch in **Preferences**, then pick a model from the gateway's available transcription models; the choice persists across launches.
+
+The switch is a two-way choice, not a fallback — the selected engine handles the whole dictation. With gateway transcription on, distinct messages cover an unreachable gateway, an unavailable model, no detected speech, and timeouts, and you can cancel an in-progress transcription from the microphone button.
+
 ## Attachments in remote sessions
 
 When you connect the mobile app to a `kilo remote` CLI session, you can share files in both directions.
@@ -85,9 +94,9 @@ Open a pull request from a PR link to review it without leaving the app:
 
 - **Overview** — PR state and CI checks at a glance.
 - **Files** — syntax-highlighted diffs with line-level comments and a file navigator.
-- **Discussion** — review threads with replies, resolve/unresolve, and reactions.
+- **Discussion** — review threads with replies, resolve/unresolve, and reactions. A **Comment on this pull request** bar at the bottom opens a composer that posts a regular PR conversation comment — it appears in the discussion right away and is not part of a review.
 
-Comments you leave are collected into a pending review on your device and submitted to GitHub as a single review. When the PR is ready, you can merge it (merge, squash, or rebase), enable or disable auto-merge, or update the branch — all from the app.
+Comments you leave on diffs are collected into a pending review on your device and submitted to GitHub as a single review. When the PR is ready, you can merge it (merge, squash, or rebase), enable or disable auto-merge, or update the branch — all from the app.
 
 PR review uses your connected GitHub account; the app asks you to connect GitHub if you have not already.
 

packages/kilo-docs/pages/customize/context/context-condensing.md

@@ -95,7 +95,7 @@ Compaction is configured in your `kilo.jsonc` file:
 
 ### Use a different model for compaction
 
-Summarization can use a cheaper or larger-context model than your main agent. Configure a dedicated compaction agent:
+Summarization can use a cheaper or larger-context model than your main agent. Pick the **Compaction model** on the **Settings → Models** tab (the Compaction section under Settings → Context links there), or configure a dedicated compaction agent in `kilo.jsonc`:
 
 ```jsonc
 {