- openclaw 2026.7.1-2 — Fixes - npm plugin updates: accept singleton-array metadata from newer npm clients so tracked official plugins can install and update to correction releases. (#108336)
- openclaw 2026.7.1-1 — Fixes - Codex progress replies: keep app-server turns running after delivered progress messages so GPT/Codex reaches its authoritative terminal response instead of stopping mid-turn. (#106961,...
- fix(memory-core): honor private-network settings for OpenAI-compatible embeddings — Fixes #94166 What Problem This Solves Fixes an issue where users selecting a configured OpenAI-compatible memory embedding provider could not consistently use its private endpoint policy. The...
- fix(test): bound direct gateway server runs — Summary - bound config-only Gateway server Vitest runs to four sequential worker processes - preserve the project's intentional non-isolated semantics inside each process - centralize Gateway server...
- Gateway server config-only tests exhaust the Vitest worker heap — Summary A config-only run of the Gateway server Vitest project accumulates every non-isolated test file in one worker until the worker reaches the V8 heap limit and exits unexpectedly. Reproduction...
- fix(memory-core): filter junk REM concept tags (issue: 111923) — Closes #111923 What Problem This Solves Fixes REM feedback-loop noise without suppressing ordinary concept vocabulary. Managed REM reflection blocks are excluded at the daily-memory ingestion...
- [AI-assisted] docs(skills): align TaskFlow bundled skills with removed runtime alias — Closes #119173 What Problem This Solves Fixes an issue where users following the bundled taskflow or taskflow-inbox-triage skills would call a runtime surface that no longer exists. Both skills still...
- fix: rebuild incomplete plugin policy refreshes — Closes #89606 What Problem This Solves A plugins registry --refresh with refreshReason: policy-changed could lose persisted path/npm install records from the in-memory plugins[] registry. A recovery...
- fix(qa): repair release validation scenarios — What Problem This Solves Release-style QA Lab validation exposed deterministic harness failures that obscured real product coverage: - the compaction retry scenario rejected the default OpenClaw...
- fix(qwen): restore the Token Plan catalog to the gateway-served set — Closes #114105 What Problem This Solves Token Plan subscribers browsing models in the picker see six rows, five of which come from the fourteen chat models their subscription actually serves. The...
- chore(deps): bump the actions group across 1 directory with 10 updates — Bumps the actions group with 10 updates in the / directory: | [github/codeql-action/init](https:/
- fix(plugins): preserve metadata reloads during models status — What Problem This Solves Fixes an issue where operators running openclaw models status could lose a newer plugin metadata publication when a config reload, plugin install, Doctor repair, or lifecycle...
- refactor(agents): propagate attribution across execution runtimes — Related: #116528 Stack: 3 of 5. Previous: https://github.com/openclaw/openclaw/pull/116793. Next: https://github.com/openclaw/openclaw/pull/116795 What Problem This Solves Resolves a problem where...
- refactor(agents): centralize immutable execution attribution — Related: #116528 Stack: 2 of 5. Previous: https://github.com/openclaw/openclaw/pull/116792. Next: https://github.com/openclaw/openclaw/pull/116794 What Problem This Solves Resolves a problem where...
- fix(ci): accept profiled partial Kova gates — Related: https://github.com/openclaw/Kova/pull/89 What Problem This Solves Fixes an issue where OpenClaw's performance workflow would reject Kova deep-profile reports after Kova correctly marked...
- refactor(agents): bind tool policy to execution attribution — Related: #116528 Stack: 4 of 5. Previous: https://github.com/openclaw/openclaw/pull/116794. Next: https://github.com/openclaw/openclaw/pull/116796 What Problem This Solves Resolves a problem where...
- fix(audit): keep execution attribution generation-aware — Closes #116528 Stack: 5 of 5. Previous: https://github.com/openclaw/openclaw/pull/116795 What Problem This Solves Resolves a problem where node execution and audit projection could correlate reused...
- fix(protocol): preserve gateway session attribution across node runs — Related: #116528 Stack: 1 of 5. Next: https://github.com/openclaw/openclaw/pull/116793 What Problem This Solves Resolves a problem where node-invoked agent runs could lose gateway-owned session...
- [Bug]: WebChat canvas hides earlier messages after main session reset — Bug type Behavior bug (incorrect output/state without crash) Beta release blocker No Summary After the main WebChat session was reset to a new session ID, the existing canvas displayed only the new...
- fix(buzz): bound NIP-11 relay information responses — Additional instructions MUST: Keep Allow edits from maintainers enabled for this PR so maintainers can help update the branch when needed. What Problem This Solves Fixes an issue where enabling Buzz...
- fix(vercel-ai-gateway): long-context usage costs are underreported — What Problem This Solves Fixes an issue where Vercel AI Gateway users would see long-context usage costs underreported when a model publishes higher prices above a token threshold. Live model...
- Session cost silently reports 0 while token usage is real; invalid configured model id silently substitutes a more expensive model — Summary Two related bugs on a long-running agent heartbeat session caused a several-hundred-dollar cost blind spot: 1. usage.cost.total is silently 0 for every Anthropic assistant message in one...
- fix(memory-wiki): scope source ownership by import mode — What Problem This Solves Importing the same physical source through both the Memory Wiki bridge and unsafe-local modes generates two source pages but records a single SQLite ownership row,...
- fix(agents): check exact provider prompts before compaction — What Problem This Solves Long tool-heavy turns could make a recovery decision from a different message view than the one actually sent to the model. That could trigger unnecessary summarization or...
- fix(cli-runner): restore history for SQLite-backed CLI resumes — Related: #98236 What Problem This Solves CLI-backed resumes could lose prior OpenClaw history after transcripts moved to SQLite. Current runtime paths pass a canonical session key in sessionFile and...
- fix: expose OAuth-backed tools to CLI agents — Closes #101967 What CLI-backed agents now carry their prepared auth snapshot through the loopback grant. Gateway tool resolution also preserves the active model context and keeps the agent credential...
- WhatsApp inbound debounce drops untrustedStructuredContext: shared contact loses name/phone when batched with a following text — What happens When a WhatsApp contact card and a text message from the same sender are merged by the inbound debouncer, the merged message keeps the placeholder in the body but loses the structured...
- [Feature]: Support env/compose configuration for gateway.controlUi.allowedOrigins in Docker/Portainer deployments — Summary Please support configuring gateway.controlUi.allowedOrigins from Docker/Compose environment variables so OpenClaw can be deployed from Portainer or other managed compose UIs without requiring...
- fix(bedrock): honor Opus 5 maxTokens when reasoning is off — Summary - Forward non-synthetic configured model maxTokens through the Bedrock Opus 5 reasoning: "off" path. - Preserve explicit per-request caps and omit OpenClaw synthetic fallback caps so Bedrock...
- fix(model-picker): preserve non-picker auth-profile override on model switch — What Problem This Solves Fixes #92244: the Telegram model picker's applySessionModelSelectionToEntry called applyModelOverrideToSessionEntry without preserveAuthProfileOverride. Switching to a...
- fix(install): block git option injection via --git-dir / OPENCLAW_GIT_DIR — What Problem This Solves scripts/install-cli.sh and scripts/install.sh both run where $repo_dir is set from the user-facing --git-dir (or OPENCLAW_GIT_DIR) argument. Git's CLI parses any positional...
- fix(telegram): duplicate inbound messages after a transient update-offset persistence failure — What Problem This Solves Fixes #105191. A transient Telegram update-offset write failure used to advance the monitor's in-memory watermark anyway. The durable watermark stayed stale, so a later...
- OpenClaw repo stats for 2026-08-04 — 385105 stars, 80945 forks, 3055 contributors
- OpenClaw sponsors (96 total) — 96 sponsors supporting OpenClaw
- fix(security): prevent package scans from exhausting worker memory (#3393) — fix(security): scope SkillSpector to bundled roots fix(security): bound bundled SkillSpector scans chore(security): format SkillSpector worker changes fix(security): import path separator for scan...
- fix: prevent worker artifact directory collisions (#3392) — fix(workers): share verified artifact materialization fix(ci): restore prepublication batch limit
- fix: upload skill files directly to Convex (#3391)
- fix: allow safe skill latest rollback (#3388) — fix: allow safe skill latest rollback chore(release): prepare clawhub 0.23.2
- fix: search prints no results when a skills.sh row matches (#3379) — Normalize skills.sh installs into the canonical downloads field so released clients can parse and render mixed-source search results.
- fix(home): contain app category scroller on mobile (#3281) — Constrain the mobile app-category strip to the available content width while preserving internal horizontal scrolling. Scope the CSS regression contract to the exact nested mobile rule.
- openclaw/openclaw on Docker Hub — Official openclaw/openclaw docker images.
- OpenClaw — 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...
- OpenClaw is a security nightmare dressed up as a daydream
- I think I found the first OpenClaw voice workflow normal people might actually keep — OpenClaw voice workflows get interesting when you stop trying to automate pizza orders and start automating boring phone errands like haircut bookings
- How to (easily) run a fully local, private AI assistant with OpenClaw and QVAC — What an agent harness is, and what OpenClaw does A language model on its own can only...
- Building an Edge-AI Computer Vision Pipeline for Real-Time Industrial Safety Interlocks — In high-risk manufacturing and process industries, optical smoke detectors and thermal threshold...
- What a good Agents.md should teach an agent on day one — I hit this last week while working inside my own OpenClaw workspace: the agent had access to the...
- ClawHub Skill Spotlight: Automation Workflow Builder — If you've ever found yourself doing the same tedious task over and over — copying data between files, checking websites for updates, generating weekly report...
- Tightening Access Rights for OpenClaw on a Synology NAS — Simplicity of POSIX vs. power of ACLContinue reading on Level Up Coding »
- Donely Review: Managed AI Agents Still Need an Adult in the Loop — The main caveat with Donely is the same thing that makes it interesting: it is designed for persistent agents that can act, not merely…Continue reading on Medium »
- I Put an AI Agent Inside My WeChat in 15 Minutes — and Gave Up Everything the Hype Promised — The pitch I kept running into went something like this: no cloud, no monthly fee, an open-source AI living inside your chat app and…Continue reading on Medium »
- Maid (AI) — Maid is a cross-platform Flutter app for interfacing with GGUF / llama.cpp models locally, and with Ollama and OpenAI models remotely.
- MiroThinker — MiroThinker is an open-source search agent model, built for tool-augmented reasoning and real-world information seeking, aiming to match the deep research experience of OpenAI Deep Research and...
- Apollo AI — Customizable interface for private, on-device AI model chatting, open source and closed source LLM connections, OpenRouter support, custom backend support, and unified access to language models...
- Loin l’idée de faire mon rageux mais je viens de recevoir ce mail de Polsia Je me suis inscrit sur le site quand il y a
- @Fei2411 openclaw 发布的时候，我不知道它到底可以为我做什么？ 到现在龙虾凉了我也不知道 — @Fei2411 openclaw 发布的时候，我不知道它到底可以为我做什么？ 到现在龙虾凉了我也不知道
- RT @Abdulaziz_Hmadi: أحد أفضل وكلاء الذكاء الاصطناعي، الشهير OpenClaw تنزله على جهازك وتربطه بأدوات الذكاء الاصطناعي مثل
- 🚨 One #WhatsApp message. Full host compromise. The #OpenClaw #WhatsAppVulnerability shows how a single malicious messag
- این روزها مصداق بارز یکی برای همه، همه برای یکی :) Vim, Ollama, n8n , OpenClaw
- @NeonNodeRH @openclaw can hermes agent mint?
- @clairevo @aaaronnte @evedev_ i tried to adopt openclaw 3 times, it never lived up to the expectations now i’m building
- RT @AISuperDomain: 🚀太强了！微软刚刚开源了 Orchard：一套面向 Agent 训练、强化学习与评测的可扩展基础设施。 它最大的价值，是让开发者可以直接在 Codex、OpenClaw、ZeroClaw 等真实 Ag
- @auspiceshirley @CyberKanjousen openclaw有个心跳检测机制。每隔一段时间就会在后台发送检测日志。估计是这个搞的锅
- RT @gregers_dev: Concordium Proof Packs can deliver OpenClaw-style verification traces — without touching the runtime.
- RT @AISuperDomain: 🚀太强了！微软刚刚开源了 Orchard：一套面向 Agent 训练、强化学习与评测的可扩展基础设施。 它最大的价值，是让开发者可以直接在 Codex、OpenClaw、ZeroClaw 等真实 Ag
- Its been so long I have even heard of openclaw. I still remember when openclaw was everywhere, hundreds of thousands o
- @michalblaha 12/ Maximálně někdo jednorázově vygeneruje nesmysl jako OpenClaw plný bezpečnostních chyb, který je následn
- Concordium Proof Packs can deliver OpenClaw-style verification traces — without touching the runtime. Goal: let counter
- RT @JulianGoldieSEO: There's a OpenClaw FULL COURSE 5 HOURS (Build & Automate Anything) https://t.co/yN4QZk2O3g
- There's a OpenClaw FULL COURSE 5 HOURS (Build & Automate Anything) https://t.co/yN4QZk2O3g
- 最近刷到 @GOATNetwork 的几条动态，感觉他们七月份干了不少实事。 ✨说白了，GOAT Network 一直在搞一件挺有意思的事：想让比特币别再只是躺着存钱或者被包装成各种 DeFi 产品，而是真正能跑智能合约、能跨链付钱、甚至 — 最近刷到 @GOATNetwork 的几条动态，感觉他们七月份干了不少实事。 ✨说白了，GOAT Network 一直在搞一件挺有意思的事：想让比特币别再只是躺着存钱或者被包装成各种 DeFi 产品，而是真正能跑智能合约、能跨链付钱、甚至让 AI 智能体自己赚钱花销，最后安全还是落回比特币本身。 https://t.co/bzvXMziPPt https://t.co/pj1fg8kKyX
- ¿Te imaginas crear un sitio web WordPress seguro simplemente hablándole a una IA? 🤯 En este vídeo ponemos a prueba Open
- @LeafShotAI Yes, I use Hermes Agent too! I’ve split the roles between them: OpenClaw analyzes soil conditions and contro
- xPressは「安いAIで並列に下書き→軽量補正→高性能AIが検証」という流れで、並列生成の弱点だった“つながりの悪さ”を補正。Qwen3-8B＋dFlashでは受理長が平均約30%改善、H200の貪欲デコードで平均約1.3倍高速化。 ht