- openclaw 2026.6.34 — 2026.6.34 Highlights - Safer browser and network boundaries: sandboxed browser routes, trusted DNS targets, custom browser origins, and loopback provider endpoints now reject unsafe access paths....
- openclaw 2026.6.33 — 2026.6.33 Highlights - Safer network and secret boundaries: provider streams, Discord REST responses, browser fetches, OAuth paths, and logs now cap hostile response sizes and keep Telegram...
- fix(agents): record stop reason in completion trajectory — Closes #118673. The embedded attempt finalizer already uses result.lastAssistant?.stopReason to classify terminal output, but omitted it from the model.completed trajectory event. This PR records...
- test(providers): cover clone-tee binary rejection cleanup — What Problem This Solves Production behavior is already fixed on main by #120351: MiniMax now reaches the shared binary-response reader, which starts rejected-body cancellation without awaiting a...
- Control UI: concurrent custom-group creation silently loses one group — Summary Creating different custom session groups concurrently from two Control UI tabs silently loses one of the groups. Environment - OpenClaw Gateway: 2026.7.2 - Control UI: 2026.7.2 - Chromium...
- main CI red since #120497: context-usage provenance sentinel regressions (compact-large-1/2) — Summary Main is red since fix(sessions): require provenance for fresh context-size facts (#120497) landed 2026-08-08T06:06Z. No CI workflow has run on main since (last success: db90dff139,...
- feat: add opt-in Nephesh dream provenance integration — Problem OpenClaw's Dream Diary contains generated scenes that an external memory system cannot safely distinguish from lived experience during ingestion. Proposed solution Add an opt-in...
- [Bug]: restored requester-settle wakes start before Gateway request scope and create a retry storm
- Control UI: render audio attachments as chat-style voice bubbles (like WeChat/QQ) — Summary In the Control UI (WebChat), audio/voice attachments from the assistant are currently rendered as a small player card (play button + duration), not as a chat-style voice bubble like WeChat/QQ...
- feat(workboard): bounded receipt-backed closeout tracker — What Problem This Solves A separate integration/activation workstream needs a bounded way to close one known outward-facing message without rebuilding a global session-completion watcher. The tracker...
- fix(model-fallback): treat empty non-GPT completions as failed candidates (#120132) — What Problem This Solves Fixes #120132. A model-fallback candidate that completed without a deliverable assistant payload could be accepted as candidate_succeeded for non-GPT model IDs. On a visible...
- fix(ci): restore deterministic main checks — What Problem This Solves Fixes four independent failures that left main CI red: Plugin SDK API baseline drift, a cold Doctor config regression test that timed out while discovering unrelated bundled...
- fix(cli): avoid missing-facing camera snap failures — What Problem This Solves Fixes an issue where users running openclaw nodes camera snap against a node with only one front- or back-facing camera could receive one photo followed by a guaranteed...
- [Feature]: image_generate should derive OpenRouter request modalities from the model's own catalog metadata, not a hardcoded ["image","text"] — Summary OpenRouter image-generation requests always send modalities: ["image", "text"] regardless of the target model; image-only models are rejected outright. image_generate should read each model's...
- fix(agents): carry complete tool args from content_block_start input — What Problem This Solves Backends that deliver the whole tool input on the content_block_start event (instead of streaming input_json_delta chunks) produced a tool-start delta with empty args: the...
- Leaf sub-agents keep sessions_yield but lose every tool that could produce the event it waits for — Summary At the spawn-depth ceiling the sub-agent tool policy strips every tool that can produce a completion event, but keeps the tool whose only purpose is to wait for one. A leaf that calls...
- fix: restore main CI after session context provenance update — What Problem This Solves Resolves the remaining full-suite failures on main after session context totals gained explicit freshness/version provenance. It also removes two new max-lines lint failures...
- fix(queue): honor steer mode for active room events — What Problem This Solves Fixes an issue where operators who explicitly set messages.queue.mode to steer still cannot use an active room-event message to correct, cancel, or unblock the in-flight...
- fix(slack): keep reasoning in one partial preview — Closes #119041 What Problem This Solves Slack users running reasoning-streaming models in partial preview mode received a new Slack message at every reasoning or tool boundary. One assistant turn...
- Gateway blocks its event loop for tens of seconds after every agent turn on multi-agent installs — Bug type Performance / runtime Summary dedupeSessionStoreTargetsBySqliteTarget resolves every session-store path's filesystem identity once per pairwise comparison instead of once per pass, so a...
- fix: iphone app voice mode issue — Closes #118474 What Problem This Solves Addresses #118474: [Bug]: iphone app voice mode issue Why This Change Was Made The implementation is intentionally scoped to the reported failure and its...
- [Bug]: AbortSignal.any() composition silently breaks signal-identity-dependent code (===, Map/WeakMap keys) — Bug type Behavior bug (incorrect output/state without crash) Beta release blocker No Summary AbortSignal.any([a, b]) always returns a new signal object, so any code that depends on signal identity —...
- feat(release): validate stable and prerelease Windows node releases — Related: openclaw/openclaw-windows-node#1104 What Problem This Solves OpenClaw release validation did not prove the exact gateway candidate against released Windows-node application artifacts. A...
- feat(browser): add relay authentication v2 — What Problem This Solves The browser relay reused a per-host relay key as Bearer, Basic, or WebSocket-token authentication. That meant the reusable key could be sent to an unverified loopback...
- Transient LLM/socket errors are retried for channels and one-shot jobs but not at the embedded-assistant stage (long turns die whole) — Summary Transient LLM/socket errors are retried for channels and one-shot jobs, but not at the embedded-assistant stage — so a long multi-step assistant turn dies whole on a single transient, while a...
- feat(gateway): persist per-origin device auth — Related: #120492 What Problem This Solves The CLI could not safely remember device credentials for a Gateway selected with --url. Operators had to keep passing a shared --token/--password, while the...
- fix: stale subagent completions are no longer delivered into a replaced requester lifecycle — Closes #118018 What Problem This Solves Fixes an issue where a parent conversation that was reset or replaced while a subagent was still running could receive the old subagent's completion in the...
- [Bug]: LF-only line endings in .cmd scripts (as written by AI editors) cause CMD to randomly drop characters / break nested blocks 鈥?silent corruption — On Windows, CMD.EXE has known fragility with LF-only (Unix newline) .cmd/.bat files. When an AI coding assistant (OpenClaw, Claude Code, etc.) saves a startup/batch script with LF endings, CMD...
- [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...
- OpenClaw repo stats for 2026-08-08 — 385525 stars, 81028 forks, 3072 contributors
- OpenClaw sponsors (95 total) — 95 sponsors supporting OpenClaw
- fix: restore homepage catalog discovery (#3418) — Restores homepage search and category discovery while preserving the canonical Trending feed contract. Adds the responsive control-group divider and closes #3417.
- fix: align discovery icons by content type and viewport (#3427) — Remove Skill listing icons across discovery, retain Plugin recognition icons on desktop, and collapse both icon columns at the existing mobile breakpoints. Keep loading skeletons aligned with settled...
- fix: restore homepage headline text selection (#3428)
- feat: add local UI proof fallback (#3429) — feat: add local UI proof fallback style: format generated Convex skills
- fix: Japanese searches skip the category and summary result tiers (#3363)
- Add skill: deciqai/first-principles (Personal Development)
- Add skill: modellix/modellix — Co-authored-by: Cursor
- 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 what Apple intelligence should have been
- 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
- 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...
- @bookwormengr don't buy it. Neither you should buy Kimi. They have slashed their quotas (just like grok slashed their im
- RT @bukanpamanmu: katanya gak perlu susah2 setup vps, openclaw/hermes dan urus api key 1. cuma ke https://t.co/zsdiUkY
- @RosoAI Tu run ca sur Claude Code / Hermes / OpenClaw ?
- #Openclaw: Ein #Prompt, ein #Bier?🍻 - https://t.co/1CKQkxLcMQ Openclaw: Ein Prompt, ein Bier? - https://t.co/1CKQkxLcMQ
- AI 时代最大的谎言，是"大家都在用 AI"。 真相是：99% 的人只是在围观别人用 AI。 OpenClaw 出来了，他们装一下，被几个报错劝退；Claude Code、Codex 火了，他们看完博主视频热血沸腾，然后死在 KYC 和 — AI 时代最大的谎言，是"大家都在用 AI"。 真相是：99% 的人只是在围观别人用 AI。 OpenClaw 出来了，他们装一下，被几个报错劝退；Claude Code、Codex 火了，他们看完博主视频热血沸腾，然后死在 KYC 和环境配置上。工具在迭代，他们在原地。 剩下那 1% https://t.co/UhGBR6RUan
- This proxy server exposes Antigravity's Claude and Gemini models via an Anthropic-compatible API for use with Claude Cod
- RT @unicity_labs: Unicity AOS installs beneath any framework you're using: @LangChain , @openclaw, without modifying a
- RT @mert: openclaw wont be a thing in 6 months
- マルチエージェント化は、成熟したAIの出口。OpenClaw🦞HermesAgent🎧を使ってわかった「AIエージェント」の本質｜ 我々の仕事は「タスクの集合」ではなく、「判断の連続」だということ。https://t.co/onfT4uHY
- 最近看到腾讯开源了一个TencentDB Agent Memory 17.7K stars, 研究了一下，分享给大家。 它主要解决了agent一个长期记忆的问题
- ここは注意が要る。游凯超がツール化を勧めたのは日付のほうで、OpenClawがツール化したのは時刻。日付はプロンプトに残したまま。少なくとも本体側では境界の後ろなので、彼が挙げたChatGPTの例と同じ置き方ではない。
- 最近看到腾讯开源了一个TencentDB Agent Memory 17.7K stars, 研究了一下，分享给大家。 它主要解决了agent一个长期记忆的问题
- @OxbfnoHome they're basically AI coding agents and CLI tools for devs. Claude Code is Anthropic's terminal agent, while
- Launched by Openclaw $AICOIN on Solana CA: CFxBZD4fvxHgvmivtob4avHFgSVDQDwYCfYX4nbezdxm https://t.co/kLQPmZfMLS https:/ — Launched by Openclaw $AICOIN on Solana CA: CFxBZD4fvxHgvmivtob4avHFgSVDQDwYCfYX4nbezdxm https://t.co/kLQPmZfMLS https://t.co/k4YzxHjZJd
- 18 capabilities & 72+ tools in one terminal: OSINT, network recon, SDR/radio, satellite, IoT, computer vision, offli
- RT @JulianGoldieSEO: There's a OpenClaw FULL COURSE 5 HOURS (Build & Automate Anything) https://t.co/3v9WDvsFh1
- Hermes と OpenClaw、よく名前が並びますが役割は別物です。 コードを書かせるなら Hermes、常駐させて雑務を任せるなら OpenClaw。どちらもモデル非依存なので、FastMetal のキー1つで両方動きます。 設定 — Hermes と OpenClaw、よく名前が並びますが役割は別物です。 コードを書かせるなら Hermes、常駐させて雑務を任せるなら OpenClaw。どちらもモデル非依存なので、FastMetal のキー1つで両方動きます。 設定の書き方と、常駐型を従量課金で回すときの注意点👇
- @pk_iv @howietl “Hyperagent is to OpenClaw what the Macintosh was to Linux” is a strong claim. Curious how that plays ou
- RT @agentnative_: People don't realize this but GPT work is basically OpenClaw in the cloud. Here is the most complete
- 明日は1日嫁氏の実家に行くので、OpenClawからローカルLLMサーバーを使えるようにセットアップしておきました。 https://t.co/VEWOcZz5MM