- openclaw 2026.4.19-beta.2 — 2026.4.19-beta.2 Fixes - Agents/openai-completions: always send stream_options.include_usage on streaming requests, so local and custom OpenAI-compatible backends report real context usage instead of...
- openclaw 2026.4.19-beta.1 — Fixes - Agents/channels: route cross-agent subagent spawns through the target agent's bound channel account while preserving peer and workspace/role-scoped bindings, so child sessions no longer...
- fix: use TextDecoder for proper GBK encoding support on Windows — Summary Fixes the Windows exec tool garbled Chinese characters issue (#56462). Uses TextDecoder from node:util to properly handle GBK encoding on Windows, which is not natively supported as a...
- feat(auto-reply): preserve TTS transcript on audio-as-voice payloads — Summary When the assistant reply is synthesized to speech — final-block auto-TTS, ACP dispatch, or the /tts audio command — the delivered payload is stripped down to {mediaUrl, audioAsVoice} only....
- Translate documentation to Chinese — Summary Describe the problem and fix in 2–5 bullets: If this PR fixes a plugin beta-release blocker, title it fix(): beta blocker - and link the matching Beta blocker: - issue labeled beta-blocker....
- fix(whatsapp): respect audioAsVoice flag in outbound delivery — Summary On the WhatsApp channel, audio replies flagged with [[audio_as_voice]] were being delivered as generic document attachments instead of native PTT (push-to-talk) voice notes. This matters...
- Skills: fix watcher so deleting a skill folder refreshes on macOS — Summary The skills hot-reload watcher used chokidar targets like /SKILL.md and //SKILL.md. On macOS with chokidar 5, removing an entire skill directory does not emit unlink for those glob-style...
- fix(memory): filter cron-triggered sessions and NO_REPLY sentinels from dreaming corpus (addresses #68449) — Addresses the second half of #68449. Companion to #68870 (stopword expansion). Problem The dreaming pipeline produces two kinds of noise, described together in #68449: 1. Concept-ranking noise —...
- fix(auto-reply): dedupe block replies on normalized media paths (fixes #68862) — Fixes #68862. Problem When a reply includes a MEDIA: tag pointing to a file outside ~/.openclaw/media/outbound/ (e.g. /tmp/foo.png or a workspace path), Telegram (and likely every channel that...
- fix(memory): expand English stopword list for concept vocabulary — Fixes #68449 Summary LANGUAGE_STOP_WORDS.english only contained 8 words (and, are, for, into, its, our, then, were), missing common articles, pronouns, prepositions, verbs, and conversation markers....
- [Bug]: mcp.servers tools not available in Pi runtime (isolated, cron, main sessions) — Bug type Behavior bug (incorrect output/state without crash) Beta release blocker No Summary MCP servers registered via openclaw mcp set are not exposed to the Pi runtime's tool_use schema. This...
- Fix Control UI image history rendering — Summary - decode transcript MediaPath and image payload variants into chat attachment content blocks - render image attachments in the shared chat UI so sent images stay visible after history refresh...
- Bug: HTTP API /tools/invoke returns 'Tool not available: browser' — Bug: HTTP API /tools/invoke returns "Tool not available: browser" Summary The browser tool cannot be invoked via HTTP API /tools/invoke, returning...
- feat(exec): inject OPENCLAW_SESSION_KEY env var for child processes — Summary Inject OPENCLAW_SESSION_KEY environment variable into child processes spawned by the exec tool, allowing scripts and skills to reliably identify the current session without relying on model...
- fix(agent): prevent fallback model from permanently overwriting agent config — Summary - Adds modelIsFromFallback flag to track when session's runtime model was selected by fallback chain - Updates resolveSessionModelRef to skip session-stored model when from fallback, ensuring...
- [Bug]: OpenRouter provider uses wrong base URL, causing all model calls to fail silently — Bug type Behavior bug (incorrect output/state without crash) Beta release blocker No Summary Version: 2026.4.15 (041266a) Platform: Linux 6.14.8-2-pve (x64) · Node.js 22.22.2 Provider: OpenRouter...
- feat: pass requesterSenderId and senderIsOwner to ChannelAgentToolFactory — Summary Pass trusted sender context (requesterSenderId and senderIsOwner) to ChannelAgentToolFactory, enabling channel plugins to implement sender-aware access control in their agent tools. Problem...
- fix(tools): enforce agent tool profile restrictions correctly — Summary - When an agent has an explicit tools.alsoAllow list, global tool sections (tools.exec, tools.fs) no longer implicitly expose those tools - This prevents a security boundary violation where...
- fix(cron): honor model override in isolated agentTurn payloads — Summary - Add regression tests verifying that payload.model is correctly forwarded to runEmbeddedPiAgent for isolated cron runs - Tests cover OpenRouter nested provider paths (e.g.,...
- fix(loader): avoid memory and compaction side effects during non-activating loads — PR 1 Submission Pack — Loader correctness fix Date: 2026-04-19 Target repo: openclaw/openclaw Related issue: openclaw/openclaw#68615 Proposed PR title fix(loader): skip plugin register for...
- fix(memory): return relevant snippets instead of file beginning in search results — Summary - Fix memory search to show relevant snippet window around matched query terms instead of always showing the file beginning - When a chunk contains matching content deep in its text, the...
- fix: extend isMinimaxVlmModel to support M2.5/M2.7/M2.7-highspeed — When MiniMax M2.7 is used as the primary model, the image tool would incorrectly send image requests directly to M2.7's chat endpoint, causing 400 errors. The image understanding for M2.7 models...
- docs(acp): clarify Claude/Codex settings inheritance — Summary - Documents how ACP Claude Code sessions inherit host ~/.claude/settings.json (model, permission mode) via HOME environment variable and the adapter's SettingsManager - Clarifies that...
- Feature: `message:enrich` hook for per-message metadata injection (e.g., inject location into each prompt) — Summary Describe the problem and fix in 2–5 bullets: - Problem: OpenClaw had no awaited pre-agent hook for injecting per-message metadata; message:received is fire-and-forget and observational only....
- fix: strip orphaned OpenAI reasoning blocks before responses API call — Summary - Wire downgradeOpenAIReasoningBlocks() into the openai-responses / openai-codex-responses stream wrapper in attempt.ts - Prevents 400 errors when conversation history contains orphaned...
- [Bug]: 400 Item 'rs_0f6f7e706597875e0069c3fc42bb7881909157c43bde349e77' of type 'reasoning' was provided without its required following item. — Bug type Regression (worked before, now fails) Summary Window11 system. Version: v2026.3.23-2 Model: gpt-5.2-chat When I call a tool using the latest version, it always show this error: 400 Item...
- Sub-agent timeout recovery creates duplicate API posts (idempotency gap) — Problem When a sub-agent makes a POST request that succeeds (HTTP 201) but the TCP connection drops before the client receives the response, the retry path re-POSTs the same content, creating...
- Active Memory plugin + qmd update chain times out at 120s on 4.15 — Summary After upgrading from 2026.3.24 → 2026.4.15, enabling the Active Memory plugin causes every blocking sub-agent call to time out, accompanied by correlated [memory] qmd update failed (Error:...
- [Feature]: Smart proxy detection - fallback to direct connection when proxy unavailable — Summary Auto-detect proxy reachability before using it, fall back to direct connection when proxy port is closed Problem to solve When a proxy (e.g., Clash Verge) is enabled, OpenClaw correctly uses...
- OpenClaw repo stats for 2026-04-19 — 360221 stars, 73423 forks, 1786 contributors
- OpenClaw sponsors (141 total) — 141 sponsors supporting OpenClaw
- enchance: mobile skills ux (#1737)
- Mobile search icon + system theme on first load — - Initialize root theme data from stored selection before paint - Hide the search label on mobile and tighten button padding
- Add branded mobile nav header — - Show the logo mark in the mobile drawer title - Tighten mobile suggestion spacing on small screens - Add test coverage for the branded mobile nav header
- Add Home link to mobile header navigation — - Insert a Home entry at the top of the mobile menu - Update header tests to cover the new menu order
- fix: harden preferences storage sync
- Update src/lib/preferences.ts — Co-authored-by: greptile-apps[bot]
- Keep diff editor mounted when switching view mode — - Remove the diff editor remount on inline vs side-by-side toggles - Add a regression test to verify the editor stays mounted
- Polish icons and brand mark styling — - Simplify home and settings labels by removing redundant icons - Swap automation icons to refresh glyphs in sidebars and toolbar - Add subtle border and shadow treatment to the brand mark
- Stabilize preference snapshots and storage sync — - Cache localStorage reads to avoid redundant snapshot churn - Sync updates across tabs and add coverage for preference re-renders
- merge: resolve AGENTS.md conflict — keep both convex-ai and stat migration rules — Co-Authored-By: Claude Opus 4.6
- style: spread footer columns evenly across full width — Switch footer grid from auto-sized centered columns to equal 1fr columns that span the full screen width. Co-Authored-By: Claude Opus 4.6
- style: adopt darker home-v2 palette globally and unify radius to 8px — Shift all theme variants (claw dark/light, hub dark/light) to the bolder home-v2 backgrounds (#060608 dark, #faf6f1 light cream). Harmonize surface, nav-bg, input-bg, and overlay-bg to match. Set...
- update: styles
- fix: restore header logo and compact/center footer — Uncomment the brand logo image in the header navbar and reduce footer vertical padding, gaps, and margins to ~55% of original height while centering the grid columns and link text. Co-Authored-By:...
- Improve about page rejection categories (#1728) — improve about page rejection categories: add icons, fix grid, polish cards - Add lucide-react icons to each rejection category card for visual scanning - Fix unbalanced grid layout by removing...
- fix: hide logo, clean up rejection categories layout (#1727) — fix: hide logo, use ClawHub as home link, and clean up rejection categories grid Comment out the brand logo image for now, rename "Immediate rejection categories" to "Rejection Categories", remove...
- docs: point morrow skill at agent-morrow namespace
- Add skill: dai-shuo/ima-knowledge-ai
- mradermacher/Qwen3.5-9b-Opus-Openclaw-Distilled-i1-GGUF — text-generation
- sulpikar2/Qwen3.5-4B-Claude-4.6-Opus-Reasoning-openclaw
- OpenClaw on DigitalOcean — Deploy your own personal AI assistant with OpenClaw on a DigitalOcean Droplet® server, a powerful open-source platform that runs entirely on your infrastructure. OpenClaw connects to the messaging...
- Tell HN: Anthropic no longer allowing Claude Code subscriptions to use OpenClaw
- Building a Systemic Autonomy Agent: OpenClaw + Gemma 4 & TurboQuant on Raspberry Pi 4B — This is a submission for the OpenClaw Writing Challenge If you’re reading this, you probably want to...
- The OpenClaw Newbie & Beyond Guide — This is a submission for the OpenClaw Writing Challenge If you follow the tech space, your timeline...
- The 90% Problem: Why Most AI Agents Are Still Broken — Building an AI agent that works is easy. Building one that doesn't break is 90% of the work. Here's what that 90% actually looks like — from leaked source code and production A/B data.
- Jentic Mini: Free API Security Layer for OpenClaw Agents — Dublin-based Jentic released Jentic Mini on March 25, 2026. A free, open-source API execution layer...
- OpenClaw v2026.3.28: xAI Grok Gets Web Search, MiniMax Brings Image Generation — OpenClaw v2026.3.28 shipped with three notable changes. xAI Grok Web Search The bundled...
- Zero Token Architecture: Why Your AI Agent Should Never See Your Real API Key — Hot take: every AI agent security guide I've read is solving the wrong problem. We spend hours...
- How I Built a Personal AI Agent That Runs on My Google Cloud VM — And Powers My Entire Portfolio
- I Built an AI Agent That Texts Me on Telegram — And It Does My Job Search
- A scheduled job should not just repeat. It should decide.
- I Stopped Treating AI Like Magic — and Built Something Useful with OpenClaw — Most people approach AI like it’s supposed to “do everything.” I tried something different: I gave...
- Integrating OpenClaw with Google Cloud: A Solo Developer's Guide — Running an AI personal assistant from your laptop sounds great until the lid closes, the Wi-Fi drops,...
- Why I’m low-key obsessed with OpenClaw (and why you probably will be too) — So I was doom-scrolling GitHub at 2am last week (don’t judge) and stumbled on this project that actually made me sit up straight. It’s…Continue reading on Medium »
- Seven AI agents had the same rule. Only one was following it — A debugging journey through placement hypotheses, fake personality audits, and a stale cache.Continue reading on Medium »
- Setting Up My Personal AI Assistant with OpenClaw — I wanted a way to automate daily tasks without juggling too many tools, so I set up a personal AI assistant with OpenClaw.Continue reading on Medium »
- Hermes Agent Without the Command Line: A 5-Minute Setup That Actually Works — Most people assume running your own AI assistant means renting a server, learning Docker, and spending a weekend in a terminal. That was…Continue reading on Medium »
- The AI Agent Skill most people miss… — First do me a huge favor, this will help me more than you think:Continue reading on Medium »
- The Self-Improving AI Agent — Hermes Agent — An Agent That Grows With You…Continue reading on Medium »
- OpenClaw Security Risks: A Warning to Users and Self-Hosters — OpenClaw 🦞 is powerful, but it currently carries serious security risk when self-hosted with default settings. Community reports and…Continue reading on Medium »
- 비개발자가 OpenClaw로 개인 AI 에이전시를 만든 5일간의 기록 — “AI가 내 일을 대신한다”는 말, 저도 처음엔...
- Building a Safer OpenClaw on a VM: Why This “Secure Stack” Repo Matters — There are plenty of ways to get OpenClaw running on a VM.Continue reading on Medium »
- OpenClaw v2026.4.9: All CLI commands hang indefinitely and get terminated with SIGKILL when called from agent exec skill
- openclaw called $ASTROID perfectly last time. running the scan again ┃🧫┃ $ASTROID ┃💰┃ MC : $1.0M ┃🎯┃ TARGET MC : $16
- just checked $MISA —openclaw showing strong upside signals ┃🧫┃ $MISA ┃💰┃ MC : $15.5K ┃🎯┃ TARGET MC : $180.6K ┃📈┃ PO
- The AI feed is not one story right now. 6 stories are colliding: Claude Opus 4.7 Grok 4.20 Multi-agent Claude Design Co — The AI feed is not one story right now. 6 stories are colliding: Claude Opus 4.7 Grok 4.20 Multi-agent Claude Design Codex Gemma 4 + OpenClaw GPT-5.4 Coding got stronger. Design got compressed....
- The AI feed is not one story right now. 6 stories are colliding: Claude Opus 4.7 Grok 4.20 Multi-agent Claude Design Co — The AI feed is not one story right now. 6 stories are colliding: Claude Opus 4.7 Grok 4.20 Multi-agent Claude Design Codex Gemma 4 + OpenClaw GPT-5.4 Coding got stronger. Design got compressed....
- just checked $assface —openclaw showing strong upside signals ┃🧫┃ $assface ┃💰┃ MC : $15.4K ┃🎯┃ TARGET MC : $208.1K
- $LUKE looking like a clean entry according to openclaw ┃🧫┃ $LUKE ┃💰┃ MC : $12.9K ┃🎯┃ TARGET MC : $153.5K ┃📈┃ POTENT
- openclaw on $XRP —momentum strong, risk low. easy ┃🧫┃ $XRP ┃💰┃ MC : $568.5K ┃🎯┃ TARGET MC : $8.6M ┃📈┃ POTENTIAL : +
- Instead of watching a 2 hour movie on Netflix, watch this masterclass on OpenClaw Course and thank me later. https://t.c
- been tracking $EVERYTHING on openclaw for a week. forecast holding strong ┃🧫┃ $EVERYTHING ┃💰┃ MC : $15.4K ┃🎯┃ TARGET
- openclaw on $281007 —signals all pointing up ┃🧫┃ $281007 ┃💰┃ MC : $36.7K ┃🎯┃ TARGET MC : $389.7K ┃📈┃ POTENTIAL : +9
- هل فكرت تشغّل نموذج ذكاء اصطناعي كامل على جهازك بدون إنترنت؟ جوجل أطلقت Gemma 4 تقدر تشغّله محليًا وتربطه مع OpenClaw
- @AdamLowisz @elonmusk @openclaw X com?
- @elonmusk @openclaw Thnx a lot
- @anuj_400 @elonmusk @openclaw Ask Grok is currently available to Premium and Premium+ subscribers only. Subscribe to unl
- been tracking $GREKT on openclaw for a week. forecast holding strong ┃🧫┃ $GREKT ┃💰┃ MC : $1.2M ┃🎯┃ TARGET MC : $13.4
- @elonmusk @openclaw @grok what's the pricing of x api
- @elonmusk @openclaw it’s like waking up on Christmas morning…
- @abhishekxbharti @Manixh02 Classic shiny object syndrome with the OpenClaw hype. People rushed Mac Minis for a private,
- openclaw flagged $MU as a strong hold. not surprised ┃🧫┃ $MU ┃💰┃ MC : $102.9K ┃🎯┃ TARGET MC : $1.5M ┃📈┃ POTENTIAL :
- openclaw confirmed what the chart was already telling me on $EVERYTHING Was Made Using - @OpenClawAIX https://t.co/ps