# Changelog All notable Metis AI releases are documented here. Release tags and GitHub releases are created locally with `pnpm release`; GitHub Actions does not publish releases. ## v1.0.24 — 2026-11-03 - Deliver files created by the agent without converting the complete file to Base64, removing the JavaScript string-size barrier even when Attachments is set to Unlimited. - Copy files asynchronously, enforce the account's saved file-size limit before and after storage, and preserve download streaming and range requests. - Recheck cancellation and the active run before registering a copied file; remove copies that could not be delivered. - Verify 630 MiB and files larger than 1 GiB with a bounded JavaScript heap, protected downloads, independent stored copies and finite account limits. ## v1.0.23 — 2026-11-02 - Transfer background file uploads in 512 KiB requests so large files pass reverse proxies with a 1 MiB request limit. - Share the request-size budget between file and voice uploads. Keep the server's existing chunk acceptance limit for older clients. - Preserve progress, retries and resuming from the server-confirmed offset; verify the complete file remains byte-for-byte identical after an interrupted upload. ## v1.0.22 — 2026-11-02 - Transfer voice recordings in 512 KiB chunks through the existing authenticated upload API, avoiding HTTP 413 responses from reverse proxies for large recordings. - Transcribe the complete reassembled recording while preserving the existing 250 MiB limit, audio validation, account isolation and provider segmentation. - Remove temporary uploads after transcription or upload failures and retain the recording for retry. ## v1.0.21 — 2026-10-28 - Distribute native installers only for Linux, macOS and Windows. Remove the container installer, image build/publishing pipeline and installation instructions. - Reject removed installation options and protect existing container installations from automatic conversion. Existing runtime detection and uninstall support remain available. ## v1.0.20 — 2026-10-28 - Set a custom Codex context window in the model options menu, with compact K/M labels and an Apply confirmation. Pass validated context and compaction settings to new and resumed Codex turns without changing the user's global Codex configuration; model and account limits still apply. - Keep reference searches responsive with background workers, a persistent normalized file index, bounded requests, cancellation and account/project access checks. - Insert note, canvas and plan references as Markdown links from the shared editor, restore internal link previews and Ctrl/Cmd-click navigation, and distinguish slash-menu actions with separate icons. - Edit note icons and upload banner images from the document toolbar while preserving draft and save-conflict handling. - Order issue board columns as Backlog, Todo, In Progress, Done and Canceled. - Forward Antigravity reasoning parameters only when the provider explicitly advertises them, and validate saved parameters against the discovered model contract. - Improve chat submission, history loading, notification polling and remote-device presence with background processing and bounded persistence work. - Match tool details, provider retry controls and automation interfaces to the selected theme. ## v1.0.19 — 2026-10-24 (updated 2026-10-26) - Introduce Notebooks with separate project and No Project file trees, nested Markdown documents, and migration of existing note hierarchies. - Add Issuebooks with issue documents, statuses, priorities, assignees, relations, comments, and agent tools for managing issues. - Unify Markdown editing across notebooks, plans, canvases and workspace files with a block editor, attachment support, and draft recovery. - Import and export notebook archives, duplicate or move documents, and open notes directly from tool results and internal links. - Improve chat search responsiveness with indexed background search, and streamline the command palette, grouped tool details and browser progress feedback. - Save theme selection and custom JSON themes per account, with import/export and synchronization between sessions. - Organize Settings into compact category overviews and detail pages, with a dedicated mobile main menu and Website/GitHub links at its bottom. - Add optional automatic provider-error retries with up to three attempts and expandable error details, while preserving provider-limit waiting and cancellation. - Let agents move chats between existing projects through account-scoped tools. - Improve local Ollama discovery, provider-account model selection, browser stream cleanup, and memory-aware installer builds. ## v1.0.18 — 2026-10-21 - Open large chats through indexed message and activity pages, load tool results and subagent histories on demand, and keep unrelated requests responsive during legacy history preparation. - Speed up note pinning with batched metadata reads, changed-only writes, inherited global pins and a searchable chat picker with fifty choices per page. - Collect server metrics in the background, retain fifteen minutes across restarts, and add detailed tooltips and one-, five- and fifteen-minute chart ranges. - Review, edit and reorder Deep Research plan steps before approval, then receive the completed report as an attached file. - Render inline source badges and Lucide chat icons consistently during streaming, including aligned wrapped text after leading icons. - Keep durable global memory guidance consistent across providers and distinguish project message authors. - Bound updater log decoding, retained output and legacy update history to avoid loading complete logs into memory. - Capture and preserve file-edit snapshots across gateway and remote-client execution for accurate diffs and rollback. ## v1.0.17 — 2026-10-20 - Open recorded file changes directly in the workspace diff editor, switch between edits, and open the current original file while preserving unsaved drafts. - Add a Deep Research mode with source search, opening, finding and auditing tools, plus a composer shortcut for switching modes. - Search references without loading full chat histories, cache bounded chat snapshots, and defer new worker jobs when host or cgroup memory is constrained. - Keep actionable account notifications synchronized with jobs and project invitations, and enable remote-device notifications by default while respecting saved preferences. - Consolidate UI guidance on UI UX Pro Max and exclude local skill installations and MCP configuration from new Git additions. ## v1.0.15 — 2026-10-19 - Invite users to shared projects, manage members and invitations, and use an account notification inbox with unread indicators while keeping provider, model and mode settings private. Automatic run-completed alerts no longer appear in that inbox. - Show edited files above the composer only while a reply is running, then keep them under the completed reply. Improve mobile note grids and preserve existing file navigation. - Preview internal chats, projects, notes and workspace links consistently from chat and workspace content. - Configure account attachment limits and keep background and queued uploads consistent with those settings. - Allow multiple Metis accounts to share an OS user and support `/` as an administrator workspace without changing the filesystem root's ownership. - Show Logout in the account menu, project members' display names and avatars, and the actual target MCP server in tool badges. - Refresh file trees and searches automatically while preserving unsaved editor changes. Preview media, Markdown and sandboxed HTML in the file editor, and inspect each recorded edit in an inline diff. - Restore clickable response suggestions across providers with shared prompt guidance, streaming delivery and persistence. - Let agent file attachments use the available message width on mobile, tablet and desktop. ## v1.0.14 — 2026-10-13 (updated 2026-10-16) - Search workspace file names and contents, export notes, plans and canvases as Markdown, plain text or PDF, and download workspace files. - Synchronize editor content with external changes, preserve unsaved drafts, detect save conflicts, search within files and highlight agent edits with links to affected lines. - Improve streamed text and Markdown animations, chat activity indicators, browser frame feedback and tool-call details, while respecting reduced-motion settings. - Preserve queued attachments across chat switches and isolate pending uploads and cached messages by chat. - Configure Remote Client notifications globally and per device, control agent-finished alerts, and expose delivery status through the notify tool while retaining compatibility with customnotify. - Recover saved form answers when a provider stream ends, prevent duplicate continuations, and allow custom responses in choice fields. - Improve Project Hub and agent-team loading, caching and error recovery; normalize Google and Vertex tool schemas for provider compatibility. - Keep Ask user forms connected beyond the default five-minute HTTP limit and preserve submitted answers for safe recovery after a provider disconnect, without starting duplicate runs. - Choose an optional default model and model parameters for projects and enforce the selection across project chats, team agents and queued runs, while keeping agent modes independent. - Open the recipient agent chat from handoff cards, move prompt and reply details into a three-dot menu, and display the assigned prompt as a normal user message in the recipient chat. - Show the queued message count next to sidebar chat actions on hover or keyboard focus, including project agent chats, with immediate updates in the active chat. - Pick custom project colors with a color picker or HEX input and search all 2,007 Lucide icons while preserving existing icons and uploaded logos. - Require valid worker run leases for internal tools, forward job and lease credentials to remote devices, enforce remote approval policies and validate legacy file-operation fallbacks. - Keep the current Skill Library view after installations and imports, and bundle Word, PDF, PowerPoint and spreadsheet skills. ## v1.0.13 — 2026-10-11 - Add Agent Hub for managing skills, connections, modes and rules, including a bundled Skill Creator, skills.sh discovery and share links for importing personal skill packages. - Open Settings and Agent Hub as account pages, add settings search, and introduce profiles with custom handles and animated avatars plus private usage dashboards and configurable API-value estimates. - Unify scoped memories across tools and workspaces, support typed interactive forms, and keep agents paused until a requested answer is submitted. - Improve automation creation, note editing, file references, notifications, chat continuity and loading older messages without losing the reading position. - Add per-user agent runtime limits and scoped stop controls, improve execution isolation and streaming, and reload the interface once after a verified successful installer update. ## v1.0.12 — 2026-10-08 - Restore the rotating disclosure arrow on grouped activity summaries such as commands, files and edits so their expand/collapse action stays visible. - Keep individual tool details in consistent popovers without disclosure arrows. ## v1.0.11 — 2026-10-08 - Manage project memories in a dedicated modal and control global memory access with “Include global memory”; project-only agents cannot read or write global memories. - Keep the current task list visible above chat, with static dots for active tasks and checkmarks for completed tasks. - Open file, edit, terminal and other tool details in consistent popovers; remove rotating disclosure arrows from tool rows and activity groups. - Animate running, queued and waiting statuses across chats and agent teams, while respecting reduced-motion preferences. - Protect note edits with serialized saves, draft recovery and explicit conflict resolution. - Improve long-chat responsiveness and synchronize project team statuses. - Pin release installers to the selected version, verify checksums and release manifests, and test an isolated Docker stack without touching live installations. ## v1.0.10 — 2026-10-06 - Add scratchpad notes with movable, resizable text, image and file cards, plus paste and full-area drag-and-drop support. - Support uploads up to 1 GiB with progress, estimated remaining time, retry and cancellation; continue uploads while switching chats or workspaces. - Share file embeds across chat, notes, plans and canvas: playable video and audio, images, PDFs, text and JSON previews, editable workspace content and removable attachments. - Stream media with range requests for playback and seeking, and move note attachments into the top toolbar. - Keep the composer mounted while the first message appears to preserve the caret and scroll position. - Add device-local microphone selection and a per-account transcription dictionary. - Keep automatically captured task context scoped to chats; save global memories only through explicit memory actions, and improve memory tool feedback. - Refresh the Metis wordmark and favicon. ## v1.0.9 — 2026-09-21 - Stabilize voice recording by keeping waveform callbacks consistent across renders. - Start the login form with an empty username and add an accessible password visibility toggle. ## v1.0.8 — 2026-09-20 - Insert voice transcripts into the focused mobile composer instead of dropping them after transcription. - Paint the recording waveform outside React so chat messages do not re-render at 60fps. - Cap client chat snapshots at 8 and skip cache writes while a run is streaming. ## v1.0.7 — 2026-09-20 - Keep interleaved text between tool groups in chat instead of collapsing every tool call to the bottom of the message. - Type `docker-urls` so the production build can import it. - Default the Linux installer to native systemd; use `--docker` for Compose. - Expose the sidebar browser settings (workspace, realtime, FPS, viewport) on Settings → General. - Fix Docker so `.env` bind/host changes apply, stop the MCP restart loop, and add `reload.sh` (`reload.ps1` on Windows) because `docker compose restart` does not recreate published ports. ## v1.0.6 — 2026-09-17 - Persist composer drafts, agent-generated chat titles, and explicit-only memories. - Let users lock chat titles so agents cannot overwrite them. - Keep the chat view stable after completion and reconcile the sending device with the durable server snapshot. - Persist automation model options, restyle the split view, and improve live chat plus remote editing. - Improve browser loading, streaming rendering, and long-chat performance. - Harden network access, production checks, native Next updates, updater health checks, version selection, and persisted update jobs. - Move release publishing to the local account-based workflow; GitHub Actions no longer publishes releases. - Fix context usage leaking across model/context-tier switches and start native sessions fresh with compacted recovery context when required. - Fix Codex runs hanging after the last visible token. - Preserve provider output after compaction. - Publish Docker, Linux/macOS, Windows, native, and source installers on the GitHub release so each user can install the way they want. - Print `You can change this. Add: /.env` under the Open URL after install. - Install Chromium with native installers and let the UI install the browser when it is missing. ## v1.0.5 — 2026-09-12 - Improved browser loading and streaming performance. - Kept Codex sessions from hanging after the last visible token. - Improved updater health checks and version handling. ## v1.0.4 — 2026-09-12 - Added the versioned updater and installer release flow. - Stabilized native Next updates and production slot replacement. ## v1.0.3 — 2026-09-12 - Improved updater and release metadata handling. - Added safer persistence for runtime jobs and sessions. ## v1.0.2 — 2026-09-12 - Improved release versioning and installer metadata. - Tightened production build and update checks. ## v1.0.1 — 2026-09-12 - Isolated updater tests from GitHub Actions environment variables. ## v1.0.0 — 2026-09-05 - Established the versioned release and updater pipeline. - Added versioned Docker and native installer assets.