OpenClaw

⚠️ Breaking change: the Qwen qwen-portal-auth OAuth integration has been removed. Migrate to Model Studio using openclaw onboard --auth-choice modelstudio-api-key

What's new in OpenClaw (v2026.3.28 → v2026.4.1):

• New /tasks chat command to view background tasks for the current session
• Built-in SearXNG web search provider plugin added with configurable host support
• Amazon Bedrock Guardrails support added
• macOS Voice Wake option added to trigger Talk Mode
• Feishu Drive comment-event flow added for document collaboration workflows
• Global default provider parameters now configurable via agents.defaults.params
• Per-job tool allowlists added for cron jobs via openclaw cron --tools
• WhatsApp agents can now react to messages with emoji
• Telegram delivery error suppression controls added to reduce noise from repeated failures
• New GLM models added to the Z.AI provider catalog
• QQ Bot added as a bundled channel plugin with multi-account support
• LINE outbound image, video, and audio sends added
• Matrix streaming replies now update the same message in place instead of sending new messages
• MCP remote HTTP/SSE server support added
• Fixed a gateway startup hang introduced in v2026.3.31 under SQLite pressure
• Fixed tasks showing stale completed entries in /status
• Fixed exec approvals not honoring security defaults in some configurations
• Fixed /model changes interrupting active runs instead of queuing behind them
• Fixed Anthropic thinking blocks being lost across compaction and context pruning
• Fixed MiniMax image generation requiring manual plugin allowlisting
• Fixed LINE channels failing to start after global npm installs on v2026.3.31
• Fixed bundled plugin runtime dependencies being lost after install or Docker builds
• Fixed memory session reindexes skipping transcripts on restart
• Fixed Discord gateway recovery and reconnect handling
• Fixed Telegram forum topic exec approvals staying in the correct topic thread
• Fixed auth profile SecretRef objects causing crashes on .trim() after upgrade
• Multiple additional stability, security, and compatibility fixes across channels and agents
• And much more. Full details at https://github.com/openclaw/openclaw/releases

⚠️ Breaking change: the Qwen qwen-portal-auth OAuth integration has been removed. Migrate to Model Studio using openclaw onboard --auth-choice modelstudio-api-key

What's new in OpenClaw (v2026.3.24 → v2026.3.28):

• xAI/Grok now uses the Responses API with built-in web search (x_search) and automatic plugin enablement — no manual toggles needed
• MiniMax image generation added for the image-01 model, supporting generate and image-to-image editing
• Plugins can now pause tool execution and ask for your approval before proceeding, via Telegram buttons, Discord interactions, or the /approve command
• ACP binds added for Discord, BlueBubbles, and iMessage so you can turn any chat into a Codex-backed workspace
• Gemini CLI backend support added alongside Claude CLI and Codex CLI
• Slack gains an explicit file upload action for channels and DMs
• Microsoft Teams and Google Chat now support unified file sends via upload-file
• Fixed infinite echo loop in WhatsApp self-chat DM mode
• Fixed long Telegram messages splitting mid-word instead of at word boundaries
• Fixed Mistral API calls failing with 422 errors
• Fixed sensitive config being accidentally exposed in the Control UI raw editor
• Fixed Discord gateway recovery looping on poisoned resume state
• Fixed GitHub Copilot token refresh falling into a hot loop on large expiry values
• Fixed memory search failing with unknown provider errors across split plugin runtimes
• Fixed iMessage leaking inline reply tags into delivered message text
• Fixed WhatsApp mentions in quoted messages falsely triggering mention gating
• Fixed Telegram verbose tool summaries not appearing in forum topic sessions
• Multiple additional stability, delivery, and compatibility fixes across Telegram, Discord, Slack, Matrix, Feishu, and more
• And much more. Full details at https://github.com/openclaw/openclaw/releases

What's new in OpenClaw (v2026.3.13 → v2026.3.24):

• OpenAI compatibility improvements: added /v1/models and /v1/embeddings endpoints for broader client and RAG tool support
• Microsoft Teams support upgraded with streaming replies, welcome cards, typing indicators, message edit/delete, and native AI labeling
• Skills panel redesigned with status-filter tabs (All / Ready / Needs Setup / Disabled), one-click install recipes, and API key setup guidance
• Tools panel now shows only what the current agent can actually use, with a live "Available Right Now" section in the Control UI
• Slack interactive replies restored with richer button/select rendering and improved setup defaults
• Discord auto-threads can now be renamed asynchronously with LLM-generated titles
• CLI gains --container flag to run commands inside a running Docker or Podman OpenClaw container
• Agent workspace files now show inline markdown previews with rich rendering for headings, tables, code blocks, and more
• Security fix: closed a mediaUrl/fileUrl alias bypass that could escape media-root sandbox restrictions
• Fixed fresh Docker installs failing before the gateway comes up
• Fixed broken channel startup where one failing channel could block all others from starting
• Fixed WhatsApp group echo suppression and reply-to-bot detection in linked-account group chats
• Fixed Telegram forum topic routing, photo dimension validation, and permanent delivery failure handling
• Fixed Discord gateway supervision crashes and timeout reply delivery
• Fixed restart sentinel so interrupted agent sessions resume correctly in the right Telegram topic or Slack thread
• And much more. Full details at https://github.com/openclaw/openclaw/releases

What's new in OpenClaw (v2026.3.8 → v2026.3.13):

• Refreshed gateway dashboard with modular views, command palette, mobile bottom tabs, and richer chat tools
• OpenAI GPT-5.4 (including updated Codex default in tests) and Anthropic Claude fast mode support with configurable session-level toggles
• Ollama, vLLM, and SGLang moved to a more modular provider-plugin architecture
• New sessions_yield tool so orchestrators can end a turn immediately and carry a follow-up into the next
• Slack agents can now send Block Kit messages through standard outbound delivery, with new interactive reply support
• UI/UX refinements across navigation, chat rendering, dashboard stability, and theme variants
• Numerous Telegram, Slack, Discord, Signal, Feishu, and browser automation reliability and delivery fixes
• Memory and session improvements including better compaction validation, continuity preservation, and transcript handling
• Performance improvements including reduced build memory usage and improved runtime stability
• Developer experience improvements including CLI help text, schema validation, plugin safety checks, and expanded test coverage
• Docker and deployment enhancements including timezone support, package updates, and improved service handling
• Multiple security fixes including Docker token leak prevention, improved SSRF/media handling, safer gateway/auth flows, and stricter request bounds
• And much more. Full details at https://github.com/openclaw/openclaw/releases

What's new in OpenClaw (v2026.3.2 → v2026.3.8):

• New backup commands to create and verify local state archives
• Configurable Talk mode silence timeout for voice interactions
• TUI now infers the active agent from the current workspace automatically
• Brave web search adds opt-in LLM Context mode for richer grounding results
• ACP provenance metadata support for tracing agent-to-agent interactions
• Spanish locale support added to the Control UI
• Context engine plugin interface for alternative memory and compaction strategies
• Persistent ACP Discord channel and Telegram topic bindings that survive restarts
• Per-topic agent routing in Telegram forum groups
• Docker multi-stage build for smaller runtime images
• Gemini 3.1 Flash-Lite added to built-in provider catalog
• Numerous Telegram, Feishu, Discord, Slack, LINE, Signal, and WhatsApp reliability fixes
• Security hardening across browser navigation, sandbox boundaries, webhook auth, and exec approvals
• And much more. Full details at https://github.com/openclaw/openclaw/releases

What's new in OpenClaw (v2026.3.1 → v2026.3.2):

• New PDF analysis tool with native support for Anthropic and Google models
• Expanded SecretRef support across all credential surfaces for safer secrets management
• Telegram streaming now enabled by default for new setups, with live preview out of the box
• Ollama embedding support for local memory search
• MiniMax M2.5 high-speed model added to built-in provider catalog
• New config validation command to catch errors before gateway startup
• Plugin SDK improvements: channel runtime access, audio transcription, session lifecycle hooks, and more
• Inline file attachments for subagent sessions
• Numerous security hardening fixes across sandbox, browser, webhooks, and prompt handling
• Reliability fixes for Feishu, Telegram, Slack, Discord, Signal, WhatsApp, and more
• And much more. Full details at https://github.com/openclaw/openclaw/releases

What's new in OpenClaw (v2026.2.26 → v2026.3.1):

• Built-in health check endpoints for improved container reliability
• Expanded Android node support: camera, permissions, health, notification actions, contacts, calendar, and motion sensors
• Telegram DM topics with per-topic configuration, allowlists, and session routing
• Discord thread sessions now use inactivity-based lifecycle controls instead of fixed TTLs
• Feishu improvements: reactions, group session scopes, reply-in-thread routing, Docx table creation, image/file uploads, and many reliability fixes
• OpenAI Responses API now uses WebSocket-first streaming by default
• Android voice screen TTS via ElevenLabs WebSocket in Talk Mode
• New tools: diffs plugin for read-only diff rendering, Feishu chat tooling, and more
• Web UI now supports German locale
• LanceDB memory supports custom OpenAI base URLs and embedding dimensions
• Cron improvements: lightweight bootstrap mode, delivery mode fixes, and localized UI
• Security hardening across Feishu webhook ingress, compaction audit, and browser sandbox
• Numerous reliability fixes for Slack, Telegram, Feishu, LINE, Signal, and more
• And much more. Full details at https://github.com/openclaw/openclaw/releases

What's new on umbrelOS:

• Redesigned onboarding with a streamlined setup flow
• Improved reliability with graceful shutdown handling and better recovery from configuration issues

What's new in OpenClaw (v2026.2.20 → v2026.2.26):

• External secrets management
• Android device support
• Multilingual stop phrases
• Security hardening
• And much more. Full details at https://github.com/openclaw/openclaw/releases

This release brings Apple Watch support, major security hardening, and numerous reliability improvements:

• Apple Watch companion app with inbox UI and notification relay
• iOS background wake via push notifications to reduce invoke failures when the app is backgrounded
• Paired device management with commands to remove or clear paired devices
• Telegram channel posts now share the same pipeline as regular messages for consistent behavior
• Cron and heartbeat delivery now correctly targets configured Telegram topics
• Gateway auth now defaults to token mode with auto-generated tokens on startup
• Security fixes for SSRF, prototype pollution, privilege escalation, path traversal, and command injection across multiple components
• Browser URL navigation now routed through SSRF-guarded validation
• Rate limiting added to control-plane write operations
• Webhook ingress hardened for Feishu and Zalo with token checks and replay protection

Full release notes can be found at https://github.com/openclaw/openclaw/releases

This release adds iOS app support, enhanced messaging platform features, and improved reliability:

• New iOS app with share extension, Talk Mode background listening, and improved onboarding with QR pairing
• Slack native streaming support for real-time message updates
• Discord interactive components can now be reusable and support per-button user restrictions
• Telegram inline button styling, improved streaming behavior, and better voice note handling
• Anthropic 1M context window support and Sonnet 4.6 model compatibility
• Enhanced subagent capabilities with better context management and announce routing
• Improved tool loop detection with progress-aware blocking
• Better cron job handling with webhook delivery modes and session continuity
• Voice call improvements including pre-cached greetings and automatic call ending on disconnect
• Security fixes for credential protection and config file handling

Full release notes can be found at https://github.com/openclaw/openclaw/releases

This release brings Discord interactive components, nested sub-agents, and comprehensive security hardening:

• Discord now supports rich interactive prompts with buttons, selects, modals, and file attachments for native interaction
• Nested sub-agents allow sub-agents to spawn their own children with configurable depth limits
• Major security improvements across webhook authentication, browser control, session handling, sandbox configuration, and content processing
• Enhanced messaging platform support including Telegram streaming fixes, WhatsApp media handling, and LINE webhook security
• Memory system improvements with better QMD collection management and Unicode-aware search
• Gateway and agent reliability fixes for session routing, context window handling, and compaction timeouts
• CLI and TUI improvements including better ANSI handling, multiline paste support, and token wrapping
• Fixed issue with the browser not coming up after a restart

Full release notes can be found at https://github.com/openclaw/openclaw/releases

This release focuses on security hardening, messaging platform improvements, and reliability fixes:

• Enhanced security across webhook authentication, browser control, session handling, and content processing
• Improved Telegram support with better blockquote rendering and reaction handling
• WhatsApp formatting improvements for bold and strikethrough text
• Better cron job reliability with fixes for scheduling, duplicate execution prevention, and error isolation
• Expanded AI provider support including Z.AI and updated MiniMax models
• Gateway improvements for message handling, authentication, and asset serving
• CLI enhancements with local timezone support for log viewing

Full release notes can be found at https://github.com/openclaw/openclaw/releases

This release brings significant improvements across messaging, AI models, and reliability:

• Added xAI Grok support as a web search provider and support for Anthropic Opus 4.6 and OpenAI Codex models
• Comprehensive Telegram improvements including better quote handling, markdown spoilers, and automatic recovery from stale thread IDs
• Web UI enhancements with token usage dashboard and agent management capabilities
• Major fixes for gateway session memory, cron scheduling reliability, and context overflow handling
• Enhanced security with skill code safety scanner and improved authentication for gateway assets
• Added more providers to the onboarding screen (Google Gemini, Ollama, xAI and more)

Full release notes can be found at https://github.com/openclaw/openclaw/releases

This release focuses on reliability, security, and developer experience improvements:

• Fully typed Telegram codebase with all remaining @ts-nocheck removed
• Major Cron enhancements: announce delivery mode, safer defaults, ISO 8601 scheduling, and deduplication fixes
• Messaging upgrades with per-channel and per-account response prefix overrides
• Web UI and TUI fixes for agents, model selection, streaming, and history handling
• Broad security hardening across channels, media handling, credentials, and voice calls
• Documentation updates, including expanded zh-CN content
• Fixed OpenAI onboarding and model selection

Full release notes can be found at https://github.com/openclaw/openclaw/releases