Patch Changes

• #2922 cd48995 Thanks @wbxl2000! - web: The timestamp under assistant replies now shows the message time instead of the work duration.
• #2922 cd48995 Thanks @wbxl2000! - web: Restyle the slash command and @ file mention menus: matched fragments are bold-highlighted in the slash menu, and long lists in both menus get a scroll fade and a draggable floating scrollbar.
• #2922 cd48995 Thanks @wbxl2000! - web: The background Bash panel now supports filtering by status, and clicking a task shows its command and output on the right.
• #2865 53909d9 Thanks @weivwang! - Cache content-hashed Kimi Web assets across reloads while keeping the app entry point revalidated.
• #2916 7475c2e Thanks @Grapedge! - Cancel an in-flight /init run together with the turn instead of letting it run to completion.
• #2911 249d8fa Thanks @7Sageer! - Fix sessions hanging on the second approval prompt and tool call results being dropped or mixed up in history when using a self-hosted OpenAI-compatible endpoint that renumbers tool call ids on every response.
• #2917 6cf315b Thanks @wbxl2000! - Fix bare URLs in chat output absorbing the CJK characters that follow them, which made the link unclickable or open a broken address.
• #2922 cd48995 Thanks @wbxl2000! - web: Fix the slash command panel staying open after switching sessions or when the composer loses focus.
• #2922 cd48995 Thanks @wbxl2000! - web: Replace the composer mode menu with mutually exclusive plan/goal pills on the left of the input area (arm via /plan or /goal, exit with ×); Swarm becomes a separate toolbar toggle.
• #2922 cd48995 Thanks @wbxl2000! - web: Restyle the work status pills above the composer with a borderless rounded look.
• #2910 eb72aeb Thanks @sailist! - Remove the 64 MiB limit on web session exports, so large sessions no longer fail with a file-too-large error when downloaded from the web UI.
• #2884 1811bd4 Thanks @liruifengv! - Fix startup banner text wrapping on narrow terminals.
• #2922 cd48995 Thanks @wbxl2000! - web: Fix $ content inside inline code spans being misrendered as inline math.
• #2899 102984a Thanks @oocz! - Fix MCP OAuth cancellation leaving an in-flight authorization waiting for its callback timeout.
• #2922 cd48995 Thanks @wbxl2000! - web: Fix the thinking-effort flyout being unreachable when selecting the last model in the subagent model list.
• #2876 5912d4c Thanks @7Sageer! - Fix repeated file-watcher errors on Windows when the workspace is a drive root (such as E:) or a UNC network share.
• #2916 7475c2e Thanks @Grapedge! - Show a clear error when forking a session while its turn is running, instead of copying a partially written turn.
• #2922 cd48995 Thanks @wbxl2000! - web: Fix forking sessions with very long histories always failing with a timeout.
• #2922 cd48995 Thanks @wbxl2000! - web: Selecting /goal from the slash menu now immediately arms a removable goal pill in the composer; typing and sending creates the goal without requiring the goal text after the command.
• #2922 cd48995 Thanks @wbxl2000! - web: Restyle the goal panel: the goal text and elapsed time move to the header, and actions become icon buttons.
• #2922 cd48995 Thanks @wbxl2000! - web: Fix CJK text immediately after a bare URL being swallowed into the link, which made the link unopenable.
• #2922 cd48995 Thanks @wbxl2000! - web: Adjust when plan mode takes effect: enabling it now arms a removable plan pill in the composer and only activates when the message is sent, matching goal mode behavior.
• #2922 cd48995 Thanks @wbxl2000! - web: Add a plan viewer panel: click a plan entry in the work bar to see the full plan, review results, and feedback.
• #2863 245e3d5 Thanks @LouisDM! - Prevent background task output from disrupting terminal pane borders.
• #2922 cd48995 Thanks @wbxl2000! - web: Show a clear top-center confirmation toast after exporting a session, and a clearer error message when the export fails because the session is too large.
• #2922 cd48995 Thanks @wbxl2000! - web: Fix the session list PR badge not refreshing after a PR is created from within a session.
• #2922 cd48995 Thanks @wbxl2000! - web: Restyle the session list PR badge as a small tag with a background.
• #2922 cd48995 Thanks @wbxl2000! - web: Unify session status display in the sidebar and stabilize session list ordering.
• #2922 cd48995 Thanks @wbxl2000! - web: Slash commands now support fuzzy search: find commands by description text, pinyin, or pinyin initials.
• #2922 cd48995 Thanks @wbxl2000! - web: Rework the subagent panel into a card grid layout with status filtering, showing in-progress and recently finished tasks by default.
• #2922 cd48995 Thanks @wbxl2000! - web: Fix server request timeouts being misreported as "cannot connect to the Kimi server".
• #2922 cd48995 Thanks @wbxl2000! - web: Restyle the todo panel as frosted cards and add a current-progress completion count.
• #2922 cd48995 Thanks @wbxl2000! - web: Increase the font size and row height of the user menu and the plan usage flyout.
• #2922 cd48995 Thanks @wbxl2000! - web: Rename the user menu's "Upgrade" entry to "Upgrade membership" and label the plan usage percentage as used.
• #2922 cd48995 Thanks @wbxl2000! - web: Add experimental automatic session title generation, with on-demand regeneration from the session list.
• #2922 cd48995 Thanks @wbxl2000! - web: Add a "sort by recent activity" option to the workspace-grouped sidebar view (switched in the view options menu); newly added workspaces now sort to the top.

Minor Changes

• #2830 ec84a6f Thanks @liruifengv! - Add an experimental fullscreen TUI mode. Set the KIMI_CODE_TUI_FULL_SCREEN=1 environment variable to enable it.
• #2700 c9bfe8b Thanks @7Sageer! - Add a configurable model pool for spawned subagents behind the secondary-model experiment (KIMI_CODE_EXPERIMENTAL_SECONDARY_MODEL=1, or the master flag): with the experiment on, the /secondary-model command or the [secondary_model] section in config.toml sets a default model or a small named pool that the main agent picks from per spawn. A lone legacy model key in the same section keeps working as the fallback default.

Patch Changes

• #2830 ec84a6f Thanks @liruifengv! - Render LaTeX math formulas ($…$ / $$…$$) in messages as Unicode formulas.
• #2855 30f56a2 Thanks @liruifengv! - Fix Ctrl+C being ignored during automatic retries of failed API requests.
• #2819 fe3cdae Thanks @7Sageer! - Fix sessions failing with a provider 400 error on every follow-up request after a turn is interrupted while the model is still thinking, on strict OpenAI-compatible providers.
• #2847 3b0936d Thanks @sailist! - Fix plain Markdown files (such as CHANGELOG.md) in an installed plugin's root directory being misidentified as skills when the plugin relies on the root SKILL.md fallback.
• #2843 c212ae9 Thanks @sailist! - Show project MCP launch targets in the workspace trust prompt, default to declining trust, and resolve fd and stty binaries to absolute paths so untrusted workspaces cannot plant bare-name executables before confirmation.
• @moonshot-ai/kimi-code-sdk contract change: WorkspaceTrustInfo.gatedMcpServers now carries structured WorkspaceTrustMcpServerInfo records (name, transport, and command/args/cwd or url) instead of plain strings, so SDK consumers rendering a trust prompt can show the full launch target.
• #2856 504e629 Thanks @pvzheroes125! - Refresh active MCP connections after OAuth credentials are added or reset.

Features

• Upgrade the experimental subagent model setting to a model pool: the [secondary_model] section can now hold a set of candidate models with descriptions, and the main agent picks from them per spawn based on the task.
• Set KIMI_CODE_EXPERIMENTAL_SECONDARY_MODEL=1 (or the master flag KIMI_CODE_EXPERIMENTAL_FLAG=1) before starting Kimi to enable it.
• Recommended setups:

  • Minimal: run /secondary-model in the TUI, or write a single default_model line in config.toml, to make every subagent run the same model by default; add force = true to pin that choice so the main agent cannot override it.

  • Declare a named pool with a one-line scenario description for each alias — the descriptions are what the main agent sees when choosing:

    [secondary_model]
    default_model = "kimi-code/kimi-for-coding-highspeed"
    
    [secondary_model.models]
    "kimi-code/kimi-for-coding-highspeed" = "Fast and cheap — good for daily refactoring, code explanation, and small edits."
    "kimi-code/k3" = "Strong at complex reasoning and deep debugging — pick it for hard problems."
    

  • See the subagent model pool docs for details.

• Add an experimental fullscreen TUI mode. Set the KIMI_CODE_TUI_FULL_SCREEN=1 environment variable to enable it.
• Support rendering LaTeX math formulas ($…$ / $$…$$) in TUI messages as Unicode formulas.

Bug Fixes

• Show project MCP launch targets in the workspace trust prompt, default to declining trust, and resolve fd and stty binaries to absolute paths so untrusted workspaces cannot plant bare-name executables before confirmation.
• Fix sessions failing with a provider 400 error on every follow-up request after a turn is interrupted while the model is still thinking, on strict OpenAI-compatible providers (e.g. DeepSeek).
• Fix Ctrl+C being ignored during automatic retries of failed API requests.
• Fix several known issues and make various refinements. See the changelog on GitHub for more technical entries.

Subagent model pool

The experimental subagent model configuration is now a model pool: list candidate models with descriptions under [secondary_model], and the main agent picks one per spawn based on the task; you can also pin a default with /secondary-model or default_model. Enable it with KIMI_CODE_EXPERIMENTAL_SECONDARY_MODEL=1. → Subagent model pool

Experimental fullscreen TUI mode

Set KIMI_CODE_TUI_FULL_SCREEN=1 to switch to a fullscreen alternate-screen UI: scrollable transcript viewport, mouse text selection, clickable links, and Ctrl-Shift-F search. → Environment Variables

LaTeX math rendering

The TUI now renders LaTeX math formulas ($…$ / $$…$$) in messages as Unicode formulas.

Also in this release: the workspace trust prompt now shows project MCP launch targets and defaults to declining trust, and untrusted workspaces can no longer plant same-named fd / stty executables; fixed Ctrl+C being ignored during automatic retries of failed API requests.

Minor Changes

#2816 ad12ad8 Thanks @liruifengv! - Show the live work progress of background subagents in the /tasks panel.

#2840 68ce3c7 Thanks @wbxl2000! - web: Image and video tool results now open in a fullscreen preview on click, with zoom support for images.

Patch Changes

#2731 437a1b8 Thanks @pvzheroes125! - Detect MCP servers that require OAuth without needing auth: "oauth" in the config.

#2840 68ce3c7 Thanks @wbxl2000! - web: Reduce UI stutter while AI responses stream in long sessions.

#2699 c0b61c6 Thanks @sailist! - Fix the token counts reported after compaction reading far below the real context size; they now match the numbers shown while the session runs.

#2810 64abebc Thanks @huangzheng2016! - Fix multi-select "Other" options so they can be deselected after being committed.

#2701 7cd6476 Thanks @sailist! - Fix multi-second freezes at startup or while idle when a large search index loads, replays, or rebuilds.

#2814 158c81d Thanks @huangzheng2016! - Show a clear error message on Windows when Git for Windows is not installed, instead of exiting silently.

#2838 e5be391 Thanks @sailist! - Close a Windows binary-planting gap in the footer git status: the git and gh commands used for the branch/dirty badge are now resolved to an absolute PATH location, so an executable planted in an untrusted workspace can no longer run before the workspace trust prompt.

#2840 68ce3c7 Thanks @wbxl2000! - web: Add hover tooltips to icon-only buttons.

#2840 68ce3c7 Thanks @wbxl2000! - web: Add breathing room around the fullscreen image preview so images no longer touch the screen edges.

#2740 01c74e9 Thanks @oocz! - Fix subagent tool changes in one session leaking into builtin profiles in later sessions.

#2840 68ce3c7 Thanks @wbxl2000! - web: Unify the fullscreen image and video previews with a shared circular close button and the same background overlay.

#2826 3c9e3b2 Thanks @liruifengv! - Page the /sessions picker list so it opens fast with large session counts.

#2840 68ce3c7 Thanks @wbxl2000! - web: Widen the sidebar's minimum draggable width.

#2723 e702817 Thanks @sailist! - Fix a spurious "Failed to steer" error when sending a message while a goal run is between turns.

#2840 68ce3c7 Thanks @wbxl2000! - web: Reduce memory and CPU usage when the app stays open for a long time.

#2825 df8ce73 Thanks @liruifengv! - Show retry progress in the loading indicator when a model request fails and is retried, with the attempt count and a detail line for the provider error.

#2840 68ce3c7 Thanks @wbxl2000! - web: Reduce memory usage and stutter during long sessions.

#2837 101c4d1 Thanks @sailist! - Remove the Agent and AgentSwarm tools from the built-in coder subagent profile, so coder subagents no longer delegate further by default. Custom profiles that list these tools explicitly can still opt in.

#2695 71ff2a0 Thanks @sailist! - Fix a Windows security risk where commands launched before the workspace trust prompt could run a malicious executable placed in the current folder.

#2813 619564d Thanks @wbxl2000! - Fix the web UI repeatedly losing its realtime connection every ~30 seconds when the server runs behind a reverse proxy or gateway with an idle connection timeout; the server now sends a WebSocket heartbeat and only closes connections that stop responding entirely.

#2842 e476c5a Thanks @wbxl2000! - Add the Modern Web Guidance plugin to the bundled plugin marketplace. Run /plugins and select Modern Web Guidance to install it.

Thanks @Leakless and @winmin for reporting the Windows binary-planting issues fixed in this release.

Features

• Add the Modern Web Guidance plugin to the bundled plugin marketplace. Run /plugins and select Modern Web Guidance to install it.
• Show the live work progress of background subagents in the /tasks panel.

Bug Fixes

• Fix coder subagents spawning further subagents by default.
• Fix the token counts reported after compaction reading far below the real context size; they now match the numbers shown while the session runs.
• Fix two binary-planting risks on Windows.
• Fix several known issues and make various refinements. See the changelog on GitHub for more technical entries.

Modern Web Guidance plugin in the bundled marketplace

Run /plugins and select Modern Web Guidance to install it and get modern web development guidance. → Plugins

Live background-subagent progress in the /tasks panel

The /tasks panel now shows the live work progress of background subagents, so you can follow along without waiting for them to finish. → Slash Commands

Also in this release: fixed the token counts reported after compaction reading far below the real context size — they now match the numbers shown while the session runs; fixed coder subagents spawning further subagents by default.

Minor Changes

• #2646 3c75a27 Thanks @liruifengv! - Show a cache-expiry reminder when resuming a long-idle session or submitting after a long idle stretch.
• #2697 e6e4ba2 Thanks @liruifengv! - web: When a model request fails and interrupts a conversation, a persistent failure card now stays in the session with one-click resume.
• #2652 68ba740 Thanks @wbxl2000! - Add Windows support for the built-in Kimi Computer Use capability and show the underlying error when capability setup fails. Install it from /plugins on Windows x64.
• #2697 e6e4ba2 Thanks @liruifengv! - web: Add a flat view to the sidebar session list.

Patch Changes

• #2648 d1ded01 Thanks @wbxl2000! - Restore how the last turn ended (completed, cancelled, or failed) when a session is resumed after a server restart, so clients can still surface a previously failed turn instead of the session looking silently stopped.
• #2666 335588e Thanks @wbxl2000! - Session listings keep how the last turn ended (completed, cancelled, or failed) across server restarts, so clients can mark previously failed sessions before they are opened.
• #2697 e6e4ba2 Thanks @liruifengv! - web: Fix the model picker overflowing the screen when many models are available, leaving the bottom options unreachable.
• #2639 8588121 Thanks @7Sageer! - /feedback now works for any signed-in user regardless of the active model; signed-out users are shown the sign-up page and GitHub Issues links instead.
• #2675 34c4181 Thanks @sailist! - Fix kimi -p exiting right after the main turn instead of waiting for background tasks and subagents to finish.
• #2694 02c026d Thanks @sailist! - Keep live sessions stable when an MCP server is removed from the workspace config or uninstalled with its plugin: its tools stay registered in open sessions but calls fail with a removal notice, and the MCP panel shows the removed status. Servers added mid-session — by a plugin install or a config edit — are not registered in open sessions; they take effect in new sessions or after /new or /reload.
• #2647 7bd3fd9 Thanks @sailist! - Read UTF-16 LE/BE text files (with or without a BOM) by transcoding them to UTF-8 instead of refusing them as binary; the web UI file viewer displays them as text as well.
• #2697 e6e4ba2 Thanks @liruifengv! - web: The sidebar error marker now only appears when the last turn failed; manually cancelled sessions are no longer flagged.
• #2697 e6e4ba2 Thanks @liruifengv! - web: Fix attachments being silently dropped when sent together with a skill command.
• #2697 e6e4ba2 Thanks @liruifengv! - web: Fix a manually chosen thinking level being reset to the model default when the first message of a new session is a skill command.
• #2697 e6e4ba2 Thanks @liruifengv! - web: Fix session renaming during IME composition — Enter no longer submits mid-composition and Esc no longer exits the editor while composing.
• #2686 ef61084 Thanks @wbxl2000! - Use a compatible PowerShell for Windows Kimi Computer Use installation, provide actionable recovery for locked plugin files, and keep its marketplace name consistent after installation.
• #2679 7b2784b Thanks @liruifengv! - web: Subagent tasks now show the model and thinking level they use.
• #2697 e6e4ba2 Thanks @liruifengv! - web: Fix text selection while renaming a session or workspace — dragging no longer moves the whole list item.
• #2697 e6e4ba2 Thanks @liruifengv! - web: Fix the chevron direction on the "show less" button of the changed-files summary card.
• #2697 e6e4ba2 Thanks @liruifengv! - web: During automatic retries after a failed model request, the working status now shows retry progress (attempt N of M) instead of looking unresponsive.
• #2677 713bf1a Thanks @liruifengv! - Fix resumed sessions rendering background task completion notifications as raw protocol text instead of a task status card.
• #2692 03aa66c Thanks @wbxl2000! - Show browser extension links and activation steps after installing Kimi WebBridge.
• #2645 2b89373 Thanks @sailist! - Fix the web UI opening the Documents folder instead of the requested file on Windows when the file path contains spaces.
• #2697 e6e4ba2 Thanks @liruifengv! - web: Fix the background-tasks and todos pills being pushed to the top of the window when the plan approval dialog expands.

Features

• web: Add a flat view to the sidebar session list.
• The Kimi Computer Use plugin now supports Windows x64 — install it from /plugins.
• Show a cache-expiry reminder when resuming or sending after a long idle. Set cache_expiry_hint to false to disable it.

Polish

• web: Subagent tasks show their model and thinking level.
• web: Show a failure card with one-click resume when a model request fails.
• web: Show retry progress (attempt N of M) in the working status during automatic retries.
• Show browser extension links and activation steps after installing Kimi WebBridge.

Bug Fixes

• Fix UTF-16 LE/BE text files (with or without a BOM) failing to load.
• web: Fix attachments being dropped when sent with a skill command.
• web: Fix the model picker overflowing the screen when many models are available.
• web: Fix a file path with spaces opening the Documents folder instead of the file on Windows.
• web: Fix the thinking level resetting to the model default when a new session starts with a skill command.
• web: Fix manually cancelled sessions showing an error marker in the sidebar; it now appears only when the last turn failed.
• web: Fix IME composition while renaming a session — Enter and Esc no longer act mid-composition.
• web: Fix dragging to select text while renaming moving the whole list item.
• web: Fix the background-tasks and todos pills jumping to the top when the plan approval dialog expands.
• web: Fix the chevron direction on the "show less" button of the changed-files summary card.
• Fix kimi -p exiting before background tasks and subagents finish.
• /feedback now works for signed-in users on any model; signed-out users see the sign-up page and GitHub Issues links.
• Fix removing an MCP server breaking open sessions: its tools stay visible but calls fail with a removal notice.
• Fix the last turn's outcome being lost across server restarts — failed turns now stay flagged in session lists and resumed sessions.
• Fix resumed sessions showing background-task completion as raw protocol text instead of a status card.

Kimi Computer Use now supports Windows x64 (Windows 10 1903+ / Windows 11) — install it from the Official tab in /plugins . → Kimi Computer Use

Cache-expiry reminder

Resuming a long-idle session or sending a message after a long idle stretch now shows a reminder that the context cache may have expired, with options to compact or start a new session first; set cache_expiry_hint to false to disable it.

kimi web stability and experience improvements

When a model request fails, a failure card now stays in the session with one-click resume; the sidebar session list adds a flat view; subagent tasks show the model and thinking level they use; and the working status shows retry progress (attempt N of M) during automatic retries.

Also in this release: browser extension links and activation steps are shown after installing Kimi WebBridge; UTF-16 text files can now be read; /feedback works for all signed-in users; fixed kimi -p exiting before background tasks and subagents finish; fixed removing an MCP server breaking open sessions (its tools stay visible, but calls fail with a removal notice).

Minor Changes

• #2407 0abcd00 Thanks @wbxl2000! - Add Kimi Computer Use and Kimi WebBridge as built-in official marketplace entries in the v2 CLI. Installing from /plugins sets up the latest managed runtime and plugin together, reports incomplete manual steps, and supports retrying interrupted setup.
• #2627 f881cdd Thanks @sailist! - Run the CLI surfaces (interactive TUI, kimi -p, kimi acp, kimi export, kimi provider) on the agent-core-v2 engine by default. Set KIMI_CODE_LEGACY_FLAG=1 to fall back to the legacy engine.
• #2565 54c04bf Thanks @7Sageer! - /fork no longer switches to the forked session: the current session stays active and its background tasks keep running. Find the fork in /sessions.
• #2630 3bd098b Thanks @liruifengv! - Ask whether to trust the current folder on startup.
• #2630 3bd098b Thanks @liruifengv! - web: Add and manage custom providers in settings.
• #2599 541ddd2 Thanks @liruifengv! - web: Overhaul the UI/UX and fix known issues.

Patch Changes

• #2601 75fe068 Thanks @wbxl2000! - Fix built-in capability availability and installed status in /plugins, preserve legacy WebBridge skills as backups during updates, and prevent Computer Use updates from duplicating or disconnecting MCP servers.
• #2635 2b3e9a9 Thanks @wbxl2000! - Rename the partner plugin marketplace tab to Curated and clarify that it contains third-party plugins from Kimi partners.
• #2614 8db7d42 Thanks @RealKai42! - Add /bug as an alias for the /feedback slash command. Type /bug to submit feedback.
• #2586 278b6af Thanks @7Sageer! - Ensure the first request waits for MCP startup to finish while the interface still opens immediately.
• #2620 2ee6e43 Thanks @xpzouying! - Fixed MCP OAuth re-authorization always failing with "Invalid redirect URI": the OAuth callback listener binds a random port per flow, but the dynamic client registration recorded the first flow's port, so every later interactive authorization was rejected at the authorization endpoint. A stale registration is now dropped automatically and the flow re-registers with the current callback URI.
• #2596 c32e661 Thanks @xpzouying! - MCP tool results now surface the spec-defined structuredContent field and _meta server metadata to the model as a serialized <mcp-structured-result> block, instead of silently dropping them. Servers that return their machine-readable contract in these fields work the same as on other MCP hosts.
• #2612 e357028 Thanks @7Sageer! - Fix all tool calls failing with spawn EBADF on macOS when a skill folder contains a very large file tree.
• #2630 3bd098b Thanks @liruifengv! - Start the interactive TUI without creating a session.
• #2630 3bd098b Thanks @liruifengv! - web: Show the signed-in account and plan usage.
• #2630 3bd098b Thanks @liruifengv! - web: Set an emoji for the session title.
• #2630 3bd098b Thanks @liruifengv! - web: Pin sessions to the top of the sidebar.

Features

• Add Kimi Computer Use and Kimi WebBridge as built-in official marketplace entries in the v2 CLI. Installing from /plugins sets up the latest managed runtime and plugin together, reports incomplete manual steps, and supports retrying interrupted setup.
• web: Add and manage custom providers in settings.
• web: Pin sessions to the top of the sidebar.
• web: Set an emoji for the session title.
• web: Show the signed-in account and plan usage.
• Add /bug as an alias for the /feedback slash command. Type /bug to submit feedback.

Polish

• Ask whether to trust the current folder on startup.
• /fork no longer switches to the forked session: the current session stays active and its background tasks keep running. Find the fork in /sessions.
• web: Overhaul the UI/UX and fix known issues.
• Start the interactive TUI without creating a session.
• Rename the partner plugin marketplace tab to Curated and clarify that it contains third-party plugins from Kimi partners.

Bug Fixes

• Fix all tool calls failing with spawn EBADF on macOS when a skill folder contains a very large file tree.
• Fix MCP OAuth re-authorization always failing with "Invalid redirect URI"; the stale client registration is now dropped and re-created with the current callback URI.
• Ensure the first request waits for MCP startup to finish while the interface still opens immediately.
• MCP tool results now surface the spec-defined structuredContent field and _meta server metadata to the model instead of silently dropping them, so servers that return their machine-readable contract in these fields work the same as on other MCP hosts.
• Fix built-in capability availability and installed status in /plugins, preserve legacy WebBridge skills as backups during updates, and prevent Computer Use updates from duplicating or disconnecting MCP servers.

Refactors

• Run the CLI surfaces (interactive TUI, kimi -p, kimi acp, kimi export, kimi provider) on the agent-core-v2 engine by default. Set KIMI_CODE_LEGACY_FLAG=1 to fall back to the legacy engine.

Official built-in plugins: Kimi Computer Use and Kimi WebBridge

The Official tab of the /plugins marketplace now lists two official built-in capabilities — press Enter to deploy the managed runtime in one step, with retry after interrupted installs and automatic backup of legacy WebBridge skills on upgrade. → Installation and Management

agent-core-v2 engine enabled by default

The interactive TUI, kimi -p, kimi acp, and other CLI surfaces now run on the agent-core-v2 engine by default; set KIMI_CODE_LEGACY_FLAG=1 to fall back to the legacy engine if needed. → Environment Variables

kimi web enhancements

The web UI now supports adding and managing custom providers in settings, pinning sessions in the sidebar, setting emoji session titles, and viewing your signed-in account and plan usage.

Also in this release: new /bug command as an alias for /feedback; startup now asks whether to trust the current folder; fixed MCP OAuth re-authorization always failing with Invalid redirect URI.

Minor Changes

• #2558 75395f6 Thanks @sailist! - Add the TurnStarted, UserPromptQueued, TaskStarted, and SessionHeartbeat hook events, enrich hook payloads with the session title and client type, include the model and profile in SessionStart, and report SessionEnd as archive when a session is archived instead of exited. Configure the new events under [[hooks]] in config.toml.

Patch Changes

• #2416 eaab2b6 Thanks @mangeshraut712! - Fall back to the built-in models.dev catalog snapshot when the public catalog is unreachable, so Known third-party provider import still works offline or in blocked networks.
• #2083 bfa0080 Thanks @StaR4y! - web: Fix dark-mode monochrome controls and align the chat composer corner radius with the design system.
• #2559 dfc55a5 Thanks @wbxl2000! - Render the "/login" already-logged-in confirmation in the success color instead of dim text, so the "Already logged in. Model configuration refreshed." message is clearly visible.
• #2572 6ba75a1 Thanks @sailist! - Rename the [loop_control] max_retries_per_step config key to max_attempts_per_step and max_steps_per_run to max_steps_per_turn: on the v2 engine the old keys no longer take effect and a startup warning prompts the rename in config.toml. The KIMI_LOOP_MAX_RETRIES_PER_STEP env var is likewise deprecated in favor of KIMI_LOOP_MAX_ATTEMPTS_PER_STEP but keeps working with a warning.
• #2585 c396873 Thanks @sailist! - Fix submitting answers to interactive question prompts being rejected when the model provider returns tool call IDs containing colons (some OpenAI-compatible gateways).
• #2562 071b6a5 Thanks @sailist! - Serve v1 message history from the server layer and drop the engine-side legacy message adapter; the /api/v1 message contract is unchanged.
• #2562 071b6a5 Thanks @sailist! - Assemble the session snapshot endpoint from the engine's services for both cold and live sessions, and remove the KIMI_SNAPSHOT_READER, KIMI_SNAPSHOT_TIMEOUT_MS, and KIMI_SNAPSHOT_CACHE_LIMIT environment knobs.
• #2563 2118544 Thanks @sailist! - Fix the context window limit showing as 0 in session status updates when no model is bound yet or the configured model no longer resolves; the limit now falls back to the default model or is omitted when unknown.
• #2563 2118544 Thanks @sailist! - The [token_counting] strategy now only selects the reported context size: estimated keeps provider-reported usage out of the context-size display, and measured no longer gets stuck retrying an oversized compaction request until it fails.
• #2563 2118544 Thanks @sailist! - Add a [token_counting] config section to choose how context token counts are derived: measured+estimated (default), measured (provider usage only), or estimated (heuristic only, for providers without usage reporting). Set strategy under [token_counting] in config.toml (or KIMI_TOKEN_COUNTING_STRATEGY) to switch.

Features

• Add four hook events: TurnStarted, UserPromptQueued, TaskStarted, and SessionHeartbeat. Configure them under [[hooks]] in config.toml — see Hooks for details.

Polish

• Rename two [loop_control] keys: max_retries_per_step → max_attempts_per_step and max_steps_per_run → max_steps_per_turn; the old keys stop working with a rename warning at startup — see loop_control.
• Add a [token_counting] config section: when a provider doesn't report token usage, switch the context-size display to local estimates — see token_counting.

Bug Fixes

• Fix answers to interactive question prompts being rejected when the model provider returns tool call IDs containing colons (some OpenAI-compatible gateways).
• Fix automatic context compaction getting stuck retrying an oversized request until it fails.
• Fall back to the built-in models.dev catalog snapshot when the public catalog is unreachable, so importing a known provider still works offline or in blocked networks.
• Fix the context window limit showing as 0 when no model is configured; it now falls back to the default model.
• web: Fix dark-mode monochrome controls and align the chat composer corner radius with the design system.
• Fix the /login already-logged-in confirmation being hard to read; it now uses the success color.