- openclaw 2026.6.8-beta.1 — 2026.6.8 Highlights - Telegram and WhatsApp channel delivery are richer and less brittle: Telegram can send structured rich text with tables, lists, expandable blockquotes, prompt-preserving CLI...
- openclaw 2026.6.7-beta.1 — 2026.6.7 Highlights - Channel delivery is tighter across Slack, Telegram, outbound media, silent replies, progress drafts, and paged action results: same-channel Slack finals persist in transcripts,...
- Codex-authenticated installs can auto-select direct OpenAI for image media understanding without OPENAI_API_KEY — Current verified scope Verified against current main on 2026-06-13. The remaining bug is the automatic image media-understanding route on Codex/ChatGPT-authenticated installs. When no explicit...
- fix(media): route OAuth image defaults through Codex — Summary AI-assisted bug fix for #87168. This PR makes implicit image media-understanding model selection auth-mode-aware for OpenAI-family defaults: - keeps implicit openai/... image media candidates...
- fix(cron): validate Telegram delivery target is numeric chat ID — > AI-assisted (Claude). All behavior proof below was produced and verified by a human-run command in a real local checkout. Summary - Problem: The cron dashboard accepted arbitrary text for Telegram...
- #92201: Embedded runner: freshly streamed thinking signatures intermittently invalid on replay (Anthropic); recovery wrapper never fires because error text is genericized — Summary Fix shouldRecoverAnthropicThinkingError to detect thinking block signature errors even when the error message has been genericized by the error-wrapping layer. The raw provider error (e.g....
- #92182: fix(slack): disable block streaming when streaming.mode=off — Summary Fixes Slack account-level streaming.mode: "off" being ignored when channel-level streaming config has block.enabled: true, causing partial streaming chunks to be posted as separate new...
- test(qa): add qa run --profile and unified output summary/evidence — Summary - Adds pnpm openclaw qa run --profile smoke-ci|release as the taxonomy-backed QA profile entry point. - Resolves profile, surface, and category selectors from taxonomy.yaml, filters mapped...
- docs: add maturity test skeleton — Summary - add a short docs/reference/maturity-tests.md page for maturity scorecard coverage - describe the current source files, scenario coverage IDs, and qa coverage workflow - link the page from...
- fix(ui): scope avatar storage per agent ID — > AI-assisted (Claude). All behavior proof below was produced and verified by a human-run command in a real local checkout. Summary - Problem: The avatar storage used a single global localStorage key...
- fix(update): preserve local package overrides — Summary - Preserves local package overrides during package update flows, including standalone added dist files, and reports the recovery bundle when overrides are not reapplied. - Keeps...
- fix(logging): prune non-idle stale diagnostic session entries — > AI-assisted (Claude). All behavior proof below was produced and verified by a human-run command in a real local checkout. Summary - Problem: pruneDiagnosticSessionStates() in...
- fix(cron): mark job active during startup catch-up replay — > AI-assisted (Claude). All behavior proof below was produced and verified by a human-run command in a real local checkout. Summary - Problem: runStartupCatchupCandidate() in...
- fix(respawn): rewrite pnpm versioned entry paths to stable wrapper (fixes #52313) — What does this PR do? When the gateway respawns after self-update via spawnDetachedGatewayProcess(), it reuses process.argv which may contain a pnpm versioned path like...
- perf: avoid event-loop stall during embedded_run bootstrap-context — Summary Fix event-loop stalls during embedded_run bootstrap-context that block the event loop for 14-22 seconds, causing message loss. Root causes addressed 1. Synchronous file I/O —...
- Runner-first / agent-action / operator Telegram delivery does not emit internal message:sent — silent hook-coverage loss for the main agent — The internal message:sent event appears to be emitted only from the auto-reply delivery layer (deliverReplies / deliverOutboundPayloads). When an agent's replies are delivered via the runner-first /...
- fix(status): warm model context-window cache before /status reads it — Problem \/status\ shows \200K\ (DEFAULT_CONTEXT_TOKENS) for any model discovered from the provider catalog, regardless of the model's actual context window. Root Cause Four independent gaps kept...
- Guard cron announce delivery against implicit last routing — Summary - reject ambiguous cron announce delivery that would resolve to implicit last routing when multiple message channels are configured - preserve deterministic routes, including...
- fix(cron): stop replaying missed slots that predate the last definition write [AI-assisted] — Summary Fixes #91944 — after a cron.update schedule change, a gateway restart (e.g. during openclaw update) could fire the job at the wrong time. The startup catch-up heuristic (isRunnableJob with...
- skills/ClawHub: surface source, add lockfile, sandbox by default — Surface source code provenance on ClawHub skill listings Summary ClawHub skill listings (web + API) currently expose no link to the underlying source code for a skill. As a user trying to verify a...
- Fold Telegram RTT sampling into live QA evidence — Summary - Rebased over openclaw/openclaw#92606 and openclaw/openclaw#92558, keeping native Vitest/Playwright QA evidence and the simplified taxonomy mapping shape intact. - Fold Telegram package RTT...
- fix(providers/openrouter): treat OpenRouter (URL + default route) as long-TTL-eligible for Anthropic cache_control — Summary isLongTtlEligibleEndpoint() decides whether the env-driven OPENCLAW_CACHE_RETENTION=long default (and conservative implicit-long behavior) promotes an Anthropic cache_control marker from the...
- Add People2.0 gate regression tests for coverage/duplicate/refund — Summary Add explicit deterministic People2.0 gate regression tests for coverage, duplicate, and refund boundary behavior in control-work. Scope - Extend control tests to cover coverage-gap,...
- AEGIS-029 Gate outputs not fully wired into queue candidate job handoff — Summary still acts as a historical queue template and is not yet formally bound to the canonical People2.0 process and gate registry introduced in AEGIS-028 follow-up work. Scope - Link the canonical...
- fix(llm): collapse cumulative openai-responses message snapshots instead of concatenating [AI-assisted] — Summary Fixes #91959 — Bedrock Mantle's openai-responses endpoint (GPT-5.x with reasoning enabled) re-emits the assistant message as many cumulative snapshot items, each a strict prefix-superset of...
- Add People2.0 machine-readable gate outcome artifact command — Summary Create a dedicated People2.0 gate-result artifact command for machine-readable audit evidence. Scope - Add a command that renders all gate outputs from import, evidence, duplicate/refund,...
- [Feature]: If HEARTBEAT.md does not exist, do not prompt the model. — Summary Skip HEARTBEAT prompt if the HEARTBEAT.md file does not exist. Problem to solve Currently, if a HEARTBEAT.md does not exist, a user issued prompt stating "Read HEARTBEAT md if it exists...
- fix(imessage): honor actions.reply=false when resolving replyToId — Summary When channels.imessage.accounts..actions.reply is explicitly false, the iMessage send path still forwarded replyToId to the imsg bridge, causing outbound delivery to fail with: The bridge...
- Fix CI — I'm preparing a PR to allow declaratively configuring Clawdis using nix, but the CI is currently red on main. This PR hopefully fixes this, so that we can have all proper checks passing on the main...
- OpenClaw repo stats for 2026-06-14 — 378617 stars, 79182 forks, 2556 contributors
- OpenClaw sponsors (110 total) — 110 sponsors supporting OpenClaw
- feat: move docs to canonical subdomain
- fix: keep proxy tests out of server routes
- fix: preserve Ask Molty auth through proxy
- fix: require npm publish proof artifact
- fix: make skill updates rollback safe
- fix: retire cli browser token callback
- fix: default cli login to device flow
- fix: require confirmation for cli token handoff
- fix: harden clawhub cli sync reporting
- fix: harden clawhub cli release and sync checks
- feat: render mermaid diagrams in markdown (#2616)
- Add skill: brewpage-publish
- openclaw-cli 2026.6.6 on Homebrew — Your own personal AI assistant
- sulpikar2/Qwen3.5-4B-Claude-4.6-Opus-Reasoning-openclaw
- mradermacher/Medina-Qwen3.5-27B-OpenClaw-Uncensored-i1-GGUF
- 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
- OpenClaw privilege escalation vulnerability
- StepFun 3.5 Flash is #1 cost-effective model for OpenClaw tasks (300 battles)
- Telegram Integration - 0$ Personal Agentic AI Assistant - Part 5 — Introduction Part 5 of the Zero Dollar AI Assistant series - Your Personal AI Assistant...
- I thought the cheap model would save my OpenClaw bill, then I watched $100 disappear in 2 days — A practical DEV post for OpenClaw users: why using one cheap model everywhere often increases total cost, how to route by task risk instead, and why c
- My AI Life OS: How I Actually Built It — The follow-up to I Gave My AI Assistant Access to My Entire LifeContinue reading on Medium »
- Giving Your AI Assistant a Voice — and Eyes — in Microsoft Teams — Giving Your AI Assistant a Voice — and Eyes — in Microsoft TeamsContinue reading on Medium »
- The AI Agent With the Keys to Your Life — What it actually takes to let an AI touch your bank account, your inbox, and your medical records — and live to tell about itContinue reading on Medium »
- OpenClaw vs Hermes: I Read 1,300 Reddit Comments So You Do Not Have To — Every few months the autonomous agent crowd finds a new holy war, and right now it is OpenClaw against Hermes.Continue reading on Medium »
- Your AI Agent Shouldn’t Be You — Why giving agents your credentials is the next major security disaster.Continue reading on Medium »
- Private GPT — Ask questions to your documents without an internet connection, using the power of LLMs. 100% private, no data leaves your execution environment at any point. You can ingest documents and ask...
- Reins — Empowering LLM researchers and hobbyists with seamless control over self-hosted models. Connect remotely, customize prompts, manage chats, and fine-tune configurations. All in one intuitive app.
- ChatSonic — Combining Google's knowledge graph, ChatSonic delivers real-time data, voice interaction, AI-produced visual content, personalized avatars, and API for services. It revisits past dialogs for informed...
- LlamaGPT — LlamaGPT is a chatbot that provides a ChatGPT-like experience, with no data leaving your device.
- Odysseus — Self-hosted AI workspace offering chat, tool use, and autonomous agent support. Enables email and research tasks securely on user hardware, keeps all data local, requires no external servers, and...
- How do I view server logs from PaperclipAI being run via "npx paperclipai run"? Seeking assistance debugging "internal server error" from curl
- https://t.co/nz5L3NI96x
- RT @Ai_Tanvir_X: Best 15 accounts to follow in AI: @karpathy = LLMs king @steipete = built openclaw @gregisenberg = s
- RT @Tanaypawar27: Best 15 accounts to follow in AI: @karpathy = LLMs king @steipete = built openclaw @gregisenberg
- RT @ClawUpAI: ClawUp entry granted! OpenClaw Summer Builder Bootcamp 2026 is now open. $5,000 prize pool + fast-track t — RT @ClawUpAI: ClawUp entry granted! OpenClaw Summer Builder Bootcamp 2026 is now open. $5,000 prize pool + fast-track to $1M in follow-on…
- Grosse update pour votre agentic : on va pouvoir communiquer plus efficacement via openclaw/hermes Ça arrive à point no
- RT @LangChainJP: 【GLM-5.2がCoding Planで利用可能に】 Z. aiは、GLM Coding Plan利用者向けにGLM-5.2を提供開始した。Claude Code、OpenClaw、Clineなど、カス
- @premium @openclaw Elon is a trillionaire, there is no need for me to give him any money a month just for some extra fea
- HE RAN OPENCLAW ON AN 8-GPU SERVER MAKING $3,400/MONTH AND JUMPED TO $12,800/MONTH WITH HERMES deploying an 8-GPU RTX 4
- "普通"不是问题，"被困住"才是。 AI 时代，普通人第一次有了杠杆。你的本事 + 一支 AI 团队 = 一门生意。 Wiseflow —— 基于 OpenClaw 的开源 AI 员工团队，GitHub 8.3K ⭐ 📖 完整长文：h — "普通"不是问题，"被困住"才是。 AI 时代，普通人第一次有了杠杆。你的本事 + 一支 AI 团队 = 一门生意。 Wiseflow —— 基于 OpenClaw 的开源 AI 员工团队，GitHub 8.3K ⭐ 📖 完整长文：https://t.co/x2gdWSMDzM #AI #搞钱 #创业 #Wiseflow
- 多くの皆様に見ていただきありがとうございます😭良かったらフォロー、いいねとリツイートよろしくお願いします。最新のAIエージェントの情報を発信しています 個人開発,ChatGPT,Cursor,Claude Code Codex Gemini
- @openclaw This mirrors our stack exactly. Cron-driven, event-triggered, human-in-loop only for exceptions.
- 感觉openclaw很少能刷到了，被抛弃了吗？
- 【GLM-5.2がCoding Planで利用可能に】 Z. aiは、GLM Coding Plan利用者向けにGLM-5.2を提供開始した。Claude Code、OpenClaw、Clineなど、カスタムモデル設定に対応するコーディン — 【GLM-5.2がCoding Planで利用可能に】 Z. aiは、GLM Coding Plan利用者向けにGLM-5.2を提供開始した。Claude Code、OpenClaw、Clineなど、カスタムモデル設定に対応するコーディングエージェントから利用するモデル。 GLM-5.2は1Mトークンのコンテキストに対応。Claude https://t.co/IUYMFEdbcn
- Want to build your own autonomous AI agent? The Starter Kit walks you through identity, memory, tools & safety. Ever
- RT @OpenAcademyAI: 🦀 OpenClaw 101 — free Duolingo-style course - 8 short lessons, theory + practice - 15 minutes and
- Le Lab X rouvre ses portes. Il y a quelques mois, j'ai monté un groupe autour de l'IA et de l'agentique. Aujourd'hui, — Le Lab X rouvre ses portes. Il y a quelques mois, j'ai monté un groupe autour de l'IA et de l'agentique. Aujourd'hui, on rouvre pour de nouveaux membres. Voilà ce qu'on a construit et pourquoi ça...
- RT @Ai_Tanvir_X: Best 15 accounts to follow in AI: @karpathy = LLMs king @steipete = built openclaw @gregisenberg = s
- RT @unegrakyatindo: Dari pada nonton bokep lebih baik anak muda belajar openclaw hanya 317 menit Potensi cuan Rp200 juta
- RT @runware: Free OpenClaw hosting, pay only for tokens Self-hosting OpenClaw means a VPS and an hour of setup. Managed
- my “wonderlands” project is now open source. this is the next iteration of my “personal AGI” project, and I’m happy to