OpenClaw changelog: what's new each month
Every stable OpenClaw release summarised by month: the highlights, new features, improvements, fixes and anything you need to act on. 5 months covered; the current month updates daily.
October 2026
17 releases: 2026.8.34 → 2026.9.36
October brought stability and security improvements across updates, recovery paths, and provider integrations. The month focused on hardening data preservation, fixing edge cases in upgrades and restarts, and adding support for new models.
Action needed
- Node runtime requirement updated: Node 24.22.1+ on the 24 line or 26.12.1+ on the 26 line required due to compromised TLS certificate key vulnerability.
- Workspace filesystem restrictions (tools.fs.workspaceOnly) now apply to Gateway-built tools, requiring approval for delegate changes and media tools.
Highlights
- Provider-native compaction now defaults for OpenAI and Anthropic, improving efficiency when primary auth profiles are unavailable.
- Gateway stays responsive during long operations and sleep/clock-jump events instead of restarting unexpectedly.
- Updates complete more reliably with better Doctor recovery, Windows/macOS platform-specific fixes, and preserved plugin state.
- Added support for GPT-6.1 Sol, Meta Muse Wonder, and MiniMax M3.1 thinking effort levels.
- Credentials and sensitive data are now masked across chunk boundaries in tool output and kept out of MCP inspection.
New
- GPT-6.1 Sol model support across OpenAI routing and Reef guard-model boundaries.
- Meta Muse Wonder model with max thinking support in catalog and onboarding.
- MiniMax M3.1 thinking effort levels and OAuth token expiry handling.
- Kimi K3.1 Pro with K3 thinking, effort, and video controls.
- Release notes formatted for people with expandable sections and plain Markdown changelogs.
- Native computer control recovery after session expiry for operator continuity.
- SQLite 3.54 data safety with automatic repair of stale full-text indexes.
Improved
- Upgrade and recovery now preserve credentials, agent state, plugin inventory, transcripts, schedules, service ownership, and runtime links.
- Plugin settings and inventory retained through migrations and pnpm package updates.
- Agent completion reliability improved with preserved delegated answers, released suspended capacity, and recovered cron output.
- Gmail, IMAP, Matrix, Telegram, remote MCP, and llama.cpp integrations repaired.
- Secret redaction of credentials in long tool output without stalling or stack overflow.
- Anthropic conversation checkpoints stay byte-identical across follow-ups, replays, and Gateway restarts.
- Gateway responsiveness improved for Control UI reconnects, model changes, dashboard work, and catalog loading.
- Interrupted sessions and automations now resume or persist after restarts.
Fixed
- Gateway keeps running after computer sleep or system clock jumps instead of restarting on stale heartbeat.
- Gateway shutdown and restart no longer hang on stalled transcript worker or in-flight session events.
- Tlon group DMs no longer treat other participants as owner, closing owner-impersonation path.
- Headless browser pages remain responsive to clicks while awaiting passkey.
- Manual /compact now settles active run before compacting instead of aborting.
- Windows and macOS package-manager updates recover from platform-specific failures.
September 2026
11 releases: 2026.8.2 → 2026.9.7
September brought major usability improvements across OpenClaw's chat, mobile, and deployment experience, with faster performance, more reliable upgrades, and support for new AI models including GPT-6 Astra. The month focused on stability, responsiveness, and making the product work seamlessly across devices and platforms.
Highlights
- Chat now displays Mermaid diagrams as visual graphics across all platforms, including native macOS, iOS, and Android apps.
- Fresh installs guide you from setup to first chat in one prompt, auto-detecting and verifying existing logins and API keys.
- Settings now apply without restarting the Gateway, except for a small set that still require restart.
- GPT-6 Astra is now available for users with OpenAI API keys or eligible ChatGPT accounts, with full image and reasoning support.
- Chat, dashboards, and session interactions stay responsive even with large transcripts and background processing.
New
- Home agent docking: open your Home agent in a right or bottom dock pane with Cmd/Ctrl+Shift+H, keeping your current page visible.
- Linux desktop companion: install on x86-64 Linux as .deb or AppImage, connect to a Gateway, and access Quick Chat from system tray or keyboard shortcut.
- Background sessions: create and run work in the background from New Session while keeping your current page open, then view completion notices.
- Personal skill libraries on shared Gateways: keep your own skills separate from workspace skills, import from ZIP archives, and share or publish per identity.
- Diagram previews with retry: Mermaid diagrams in chat can be enlarged and automatically retried if rendering fails.
- Android UI parity: chat screen, sidebar navigation, and appearance settings now match the web Control UI, with a six-line composer and optional voice recording.
- Public conversation sharing: explicitly publish a revocable read-only link to a session's existing and future conversation text.
- Searchable meeting library: browse saved notes, search transcripts, download Markdown or JSONL archives, and manage capture sources from the Control UI.
Improved
- Upgrades now roll back automatically if post-update checks fail, and preserve your configuration and secrets across failures.
- Browser control works without a running Gateway for supported macOS and Linux Chrome extension builds, waking the paired local relay.
- Four new UI themes added: CRT, Manuscript, Rosé, and Miami, with choices preserved offline and applied without flashing.
- Voice output now keeps internal reasoning out of speech, preserves tool-generated audio through delivery, and keeps browser Talk turns working after setup.
- Agent replies now return a final answer after tools complete and surface failures after accepted turns, fixing conversations stuck at tool output.
- Gateway startup now recovers under load with large agent rosters, and malformed legacy cron rows are quarantined instead of blocking boot.
- Codex tool approvals are now durable across sessions, follow the session's security posture, and reach your chat when delegated agents propose changes.
- Provider and channel safety improved: untrusted responses are bounded, oversized inputs rejected before expensive work, and malformed payloads handled safely.
Fixed
- Replies now survive Gateway restarts without losing recovery markers or discarding queued and delegated replies.
- Git backups now preserve complete text containing embedded NUL characters and support Nix-managed config and credential links.
- Browser automation and workspace reads reject malformed or oversized inputs before interrupting active agent runs.
- Local model discovery, speech, meeting, and collaboration integrations make bounded requests and surface failures without destabilizing the Gateway.
- Discord, Matrix, Telegram, Slack, WhatsApp, LINE, Feishu, Zalo, and meeting plugin edge cases that could lose, corrupt, or stall messages are repaired.
- Doctor plugin registry now preserves the complete bundled plugin inventory when creating or repairing registry state.
August 2026
5 releases: 2026.7.1-1 → 2026.8.1
August brought major usability and security improvements to OpenClaw, including new conversation search, cross-device session support, and interactive widgets in chat. The month also focused on hardening credential handling, channel reliability, and long-running agent stability.
Highlights
- Search past conversations by exact words or phrases and reopen matching messages.
- Run work on paired devices or cloud workers and move sessions with warm machines.
- Answer agent questions through interactive cards with free-text alternatives and skip options.
- Use interactive widgets in chat and pin them to session dashboards with export to images.
- Request credentials through masked prompts without exposing values in chat or model context.
New
- Find past conversations: search visible conversation text and reopen surrounding messages from matches.
- Sessions beyond your Gateway: run work on paired devices or cloud workers and move session workspace.
- Session progress card: follow durable progress through reloads and track subagent activity.
- Answer structured questions: respond through web and native cards, messaging buttons, or plain text.
- Interactive results and dashboards: use widgets in chat, pin to dashboards, grant specific actions, and export as images.
- Private credential requests: let agents request credentials through masked prompts without exposing values.
- Approve recurring work once: grant automation permissions for exact operations and inspect or revoke later.
- Richer media handling: keep audio and video attached across uploads, replies, playback, and reloads.
Improved
- Codex progress replies now let GPT/Codex reach authoritative terminal responses instead of stopping mid-turn.
- Long-running agents distinguish genuine stalls from active long model calls and wedged backends.
- Discord reconnects no longer silently drop queued messages or repeat ambiguous sends.
- Telegram credentials are preserved during service restarts through SecretRef backing.
- Extended-stable updates allow selecting and receiving availability notices for the extended-stable channel.
- MCP loopback clients now use short-lived session-bound attach grants instead of mutable child-process authority.
- Browser and network boundaries reject unsafe access paths for sandboxed routes and custom origins.
- Pending channel work resumes after recovery with idempotent acknowledgements and bounded Discord gateway bursts.
Fixed
- Memory Core startup no longer traps Gateway in fatal restart loop from derived legacy-index and cache-sidecar conflicts.
- WSL state permissions tolerate read-only errors only when existing state is already private.
- Managed plugin updates recover stale npm lock metadata for clean official plugin updates.
- Legacy migration residue remains nonfatal during startup instead of blocking healthy upgrades.
- Codex app-server commands now require actual human or plugin approval with exec auto-review bound to exact resolved command.
- SQLite checkpoints, workspace reads, and plugin HTTP responses no longer turn transient host conditions into permanent failures.
July 2026
1 release: 2026.7.1
July brought a major overhaul to OpenClaw's interface and setup experience, alongside substantial improvements across all official apps and integrations. The month focused on making the product easier to use from first installation through daily work.
Highlights
- Control UI redesigned for better conversation organization and side-by-side workflow, with clearer task tracking and cost views.
- Guided setup process now checks connections before saving and preserves your choices, getting you to your first chat faster.
- All official apps (iOS, iPadOS, Android, macOS) received major updates to setup, navigation, chat, voice, and file handling.
- Telegram, Slack, and Discord integrations improved significantly with better progress tracking, threading, and reliability.
- Expanded model support including GPT-5.6, Tencent Hy3, and Meta Model API with Muse Spark 1.1.
New
- openclaw attach command gives Claude Code temporary access to a selected session.
- Codex delegation and native subagents now return tracked results more reliably.
- Telegram integration now supports live progress, photos, documents, and topics.
- Slack integration improved with threads and card-based formatting.
- Discord integration now handles replies and voice sessions.
Improved
- Setup flow validates connections before saving and remembers your earlier choices.
- Chat controls and usage views made clearer in the main UI.
- File handling, downloads, and approvals workflow streamlined.
- Telegram delivery, account routing, and command handling improved.
- Slack conversation handling reduces unnecessary waits for longer chats.
- Discord reconnection behavior and session-resume reliability enhanced.
- Mobile app permissions and localization support improved across iOS, Android, and macOS.
- Offline reading capability added to mobile apps.
Fixed
- Discord multi-account behavior now works correctly.
- Repeated session-resume failures on Discord resolved.
- Codex-connected agents now track and return results reliably.
- Slack duplicate prevention improved.
- Discord unread cues and recovery handling corrected.
June 2026
6 releases: 2026.6.5 → 2026.6.11
June brought major reliability improvements across channels and providers, with better handling of model reasoning, safer session recovery, and expanded platform support. The month focused on making agent interactions more stable and predictable across Telegram, iMessage, Matrix, and other integrations.
Highlights
- Model reasoning and internal scaffolding are now hidden from channel output, showing users only final answers.
- Telegram, iMessage, and Matrix channels are significantly more reliable through improved context retention and message routing.
- Parallel web search is now bundled as a first-class search provider with full onboarding support.
- Extended thinking and session recovery now survive restarts and connection issues more gracefully.
- New provider support includes Claude Haiku 4.5, GLM-5.2, and Codex Hosted Search alongside improved OAuth routing.
New
- Parallel bundled web-search plugin with registration and guarded endpoint handling.
- Matrix voice-message preflight and thread-aware read/reply behavior.
- OpenRouter OAuth and Claude Fable 5 model support.
- /fast auto mode for quick conversational starts with automatic fallback to normal mode.
- Standalone official provider plugins as first-class npm releases with external installation.
- StepFun provider available from npm and ClawHub.
- Codex Hosted Search as a key-free provider option.
- iOS Watch controls and Android chat context display.
Improved
- MCP results normalized to prevent session poisoning and provider conversion failures.
- Safer channel output filtering for reasoning and internal markup.
- Better Matrix and Vertex reliability with context retention across voice notes and threaded conversations.
- Telegram and WhatsApp delivery with richer formatting including tables, lists, and blockquotes.
- Faster first replies with lazy slash-command work and cached metadata.
- Control UI startup no longer waits on broad model loading.
- More consistent fast-mode timing and progress visibility across fallback and model switches.
- Account-scoped routing for Telegram DMs, iMessage transport recovery, and improved echo markers.
Fixed
- Tighter security boundaries around transcript, sandbox, MCP, and exec-approval paths.
- Reliable Telegram delivery with streamed text surviving tool calls and maintaining coherent callbacks.
- iMessage connection survival across restarts and idle periods with durable echo markers.
- Google Chat direct messages no longer incorrectly treated as group conversations.
- Telegram quoted replies now stay attached to the correct conversation and user question.
- QQBot group admins can now control slash-command visibility and privacy scoping.
Summaries are written automatically from the official release notes (full changelog ↗); check the original notes before relying on a detail. OpenClaw: pricing, features and alternatives · All changelogs