Changelog

Track every improvement to OpenClaw. New features, fixes, and platform updates.

v2026.9.4 feature

Plugin discovery, Skill Workshop, and more control over active sessions

  • The Plugins page combines installed entries and ClawHub listings in one search and browsing surface, while chat capability questions can surface official catalog listings with current install state and the existing review path
  • Skill Workshop turns past conversations into reusable skills through a chat that can be steered
  • GPT Image 2.5 support adds another image-generation option alongside controls for cloud sessions and interactive questions in the terminal
  • Updates, memory, and messaging receive fixes across the release
v2026.6.35 fix

Extended-stable boundary hardening and delivery recovery

  • Provider and channel adapters bound untrusted response bodies and reject oversized inputs before expensive work begins
  • Agent, Gateway, retry, and channel paths preserve safe recovery across cancellations, partial sends, process-stream failures, and transport errors
  • Bundled local-model, browser, media, and collaboration plugins recover cleanly from malformed payloads, timeouts, and transient upstream failures
  • Browser automation, local tools, workspace reads, and plugin metadata handling reject malformed or oversized inputs before they interrupt an active run
v2026.9.3 feature

OpenClaw 2026.9.3: updates, prompt caching, and Skill Workshop

  • Core and plugin updates rehearse in isolated candidate state before activation, support eligible 2026.9.2 migrations, and recover abandoned update records without stopping a healthy matching Gateway
  • Warm prompt caches are preserved while cold session updates and memory search do less unnecessary work, and worker builds are reused between sessions
  • Skill Workshop keeps skills in a persistent agent-owned collection across workspaces, compares complete skill instructions, and retires missing-draft suggestions safely through Doctor
  • Selected sessions can be shared through revocable read-only links that expose existing and future conversation text while excluding tools, reasoning, files, images, and executable widgets
v2026.9.2 feature

GPT-6 Astra, live setting updates, and restart recovery

  • GPT-6 Astra is available through eligible OpenAI API-key and ChatGPT/Codex accounts, with text and image input, Responses tool calls, and supported reasoning controls
  • More agent, model, tool, channel, browser, node, access, and terminal settings apply through their running owners; settings that still need a Gateway restart remain documented in the configuration reference
  • Interrupted active, queued, and delegated replies recover after Gateway restarts while preserving continuation instructions through compaction and retry attempts
  • Automatic updates retain active settings, enabled skills, and default-agent ownership, then provide actionable recovery guidance when a Gateway restart is needed
  • Shared Gateways now default session tools to all-session visibility; operators should set tools.sessions.visibility to agent or self when they need narrower access
  • Gateway latency diagnostics break RPC latency into admission, queue wait, handler, and first-response phases in OpenTelemetry and Prometheus output so slow dispatch and execution are easier to distinguish
v2026.9.1 feature

Mermaid in chat, faster setup, and safer updates

  • Completed Mermaid blocks render as diagrams in the Control UI and native macOS, iOS, and Android apps, with previews, enlargement, source and copy controls, and mobile render retries
  • Quick Start can reuse verified Claude Code or Codex access or an existing provider key, then open the web dashboard from a foreground Gateway while Custom setup remains available
  • Personal skill libraries can sit beside workspace skills, import from ZIP archives, and be shared or published per identity on team Gateways
  • openclaw update rolls back the npm candidate after a failed post-update Doctor check, preserves configuration and secret references, waits for plugin readiness, and supports recovery when no service manager exists
  • Gateway startup recovers more cleanly under load and with large agent rosters, while malformed legacy cron rows are quarantined instead of blocking boot
v2026.8.2 feature

Linux desktop companion, docked Home, and safer updates

  • The Linux companion ships as .deb and AppImage packages for x86-64 systems, connects to local or remote Gateways, and opens Quick Chat from the system tray or an X11 shortcut
  • Home can dock beside the current page, retain the work-context snapshot under operator control, and attach selected text to the conversation
  • New Session can start background work without leaving the page while retaining its selected local, cloud, or paired-device placement
  • Update and migration paths preserve newer valid configuration, stop incomplete migrations before success is claimed, and recover a stopped Gateway only when the installed package or rollback is verified safe
  • Supported browser extension builds can wake a paired local relay for authenticated CDP clients, and the Control UI adds CRT, Manuscript, Rosé, and Miami themes
v2026.8.1 major

OpenClaw 2.0: cloud sessions, durable work, and migration tooling

  • Sessions can run on paired devices or cloud workers, move their workspace, and reuse warm machines and project seeds for later cloud runs
  • Durable progress cards retain session status through reloads and surface subagent activity and accumulating edits across web and native chat
  • Masked credential requests and an opt-in proxy keep protected-secret substitution scoped to approved destinations
  • OpenProse and OpenAI route migrations use openclaw doctor --fix, while the release keeps existing source files and flags conflicts for operator repair
  • Explicit permission modes anchor restricted filesystem access to the recorded workspace or worktree, with inherited policy for newly created worktrees
v2026.6.34 fix

Extended-stable hardening for Gateway boundaries and recovery

  • Sandboxed browser routes, trusted DNS targets, custom browser origins and loopback provider endpoints reject unsafe access paths
  • Retained session writes, provider fallbacks, stream progress handling and stdio failures recover without silently ending active work
  • Pending channel work, acknowledgements, delivery evidence, Gateway queues and overload handling recover more cleanly across retries and restarts
  • SQLite checkpoints, workspace reads, process probes, plugin responses and dependency handling tolerate expected transient host failures
v2026.6.33 feature

Extended-stable channel and long-running agent repairs

  • Package installations can select and update from the extended-stable channel without silently falling back to another release line
  • Run release, liveness checks and watchdog behavior distinguish genuine stalls from active long model calls and wedged backends
  • Provider streams, browser fetches, OAuth paths and logs cap hostile response sizes, while Telegram credentials stay out of diagnostics
  • Discord reconnects, Telegram reply handling and SecretRef credential recovery preserve queued work, intended threads and valid profiles more reliably
v2026.7.1-2 fix

Official npm plugin update compatibility fix

  • Official tracked npm plugins now accept singleton-array metadata emitted by newer npm clients, so correction releases can install and update normally
v2026.7.1-1 fix

Gateway startup and managed plugin update repairs

  • Codex app-server turns continue to the authoritative terminal response after a progress message is delivered
  • Memory Core recovers derived legacy-index and cache-sidecar conflicts without trapping the Gateway in a fatal restart loop, while structural vector-store corruption remains retryable
  • Guarded WSL state-permission repairs tolerate EROFS only when the existing state path is already private, preserving fail-closed handling for broad permissions
  • Reviewed migration residue no longer blocks otherwise healthy Gateway startup, and stale npm lock metadata no longer prevents official managed plugin updates
v2026.7.1 feature

Control UI, onboarding and native app updates

  • The Control UI now organizes conversations and parallel work around clearer session navigation, resizable panes, a live Tasks view, chat controls, usage details and Gateway health
  • Guided onboarding gives users clearer steps to a working first chat, checks connections before saving them, and preserves earlier choices when setup is interrupted
  • Official iOS, iPadOS, Android and macOS apps received broad updates across setup, navigation, chat, voice, permissions, files, scheduled work, offline reading, queued sends and connection recovery
  • Provider support expands with GPT-5.6 compatibility, Tencent Hy3 and Meta Muse Spark 1.1, while connected coding-agent and Codex workflows gain stronger session continuity
  • Repeatedly failing Gateways now leave a stable repair path rather than restarting indefinitely; scheduled work, remote browser control and workspace terminals also received updates
v2026.6.11 feature

Google Chat DM fixes and steadier conversation routing

  • Newer Google Chat direct messages no longer get mistaken for group conversations, so one-to-one chats reach the correct DM session while existing Space and group routing stays unchanged
  • Feishu voice replies now show clip duration in the chat bubble before playback starts
  • Discord and Telegram replies, plus mirrored chat history, stay attached to the intended conversation more consistently across repeated replies and session changes
  • Background image, video and music results now return to the chat that requested them even when the task starts without a full conversation target
v2026.6.10 feature

Fast talk mode, model routing and trusted policy fixes

  • Talk sessions can enable fast mode for short conversational turns, then return to normal mode for longer runs with bounded fallback and delivery behavior
  • Fast-mode state now survives retries, fallback transitions, progress events and embedded, CLI and ACP normalization paths
  • Zai model synthesis, GLM overload failover and native reasoning-level selection now follow the active model catalog more consistently
  • Channel switches reset stale origin fields, while cron delivery awareness stays attached to the target session
  • Composed hook registries keep the trusted tool policies required by approval-sensitive flows, and provider plugin onboarding refreshes registry metadata after setup installs
v2026.6.9 feature

Telegram delivery, Codex integration and provider plugin releases

  • Telegram now sends rich HTML, preserves rich markdown and sticker paths, renders progress drafts and command output more faithfully, and keeps mentions and spooled handlers on the right delivery path
  • Interrupted agent runs recover more reliably across retries, terminal outcomes, usage after compaction, session history repair and reply reconciliation
  • Codex gained automatic plugin approvals, GPT-5.3 Spark OAuth routing, remote-node exec as a dynamic tool, and more reliable app-server teardown and terminal outcomes
  • Official provider packages became standalone npm releases, externally installed channel plugins load at Gateway startup, and StepFun is available from npm and ClawHub
  • The Control UI added a session workspace rail and extension health, while search and skill installs gained Codex Hosted Search, deliberate key-free provider opt-ins and verified source provenance

Want the full technical details? Check out the GitHub releases.

View all releases on GitHub →