- OpenClaw 2026.8.1-beta.3 — Highlights - GPT-5.6 Sol, Terra, Luna, and Ultra reasoning support across OpenClaw and the Codex runtime. - Control UI first-run setup now continues verified model setup into Custodian and optional...
- [AI-assisted] docs(browser): teach the code-mode global call, not the removed tools.call API — What Problem This Solves The browser-automation skill's Code Mode Loop still taught the guest to call the Browser tool via tools.call("openclaw:browser:browser", input) and read the .result.details /...
- fix: explain cloud worker enrollment failures behind reverse proxies — Fixes #128583 What Problem This Solves Fixes an issue where operators dispatching a cloud worker through a reverse-proxied Gateway could wait through the full enrollment deadline and receive only...
- perf(ui): batch board metadata lookups and coalesce duplicate dashboard reads — Closes #125313 What Problem This Solves On a Gateway with many sessions, the Control UI issued one full board.get per visible session merely to learn whether a board exists, duplicated the mounted...
- Tool error warnings appear after assistant response instead of before/inline — Description When a tool (like edit) fails during a run, the error warning message appears after the assistant's response text, which is confusing for users. The error happened during the run, but the...
- docs(browser): expose batch/doubleClick/scrollIntoView/labels in tool schema and docs — What Problem This Solves The browser runtime already supports batch, scrollIntoView, doubleClick, and labels/annotations, but an agent reading the browser tool schema, the bundled browser-automation...
- fix(updater): protect the live gateway during clean builds — Closes #125305 What Problem This Solves Fixes an issue where operators updating a source-checkout gateway could lose the running build when scripts/update-gateway.sh cleaned dist while the gateway...
- [Bug]: tui-pty-harness e2e tests flake under CPU load (2s OUTPUT_TIMEOUT_MS) — > AI-assisted: investigated and written with Claude Code (Opus 5). The reproduction below was run end-to-end on real hardware; timings and counts are measured, not estimated. Summary...
- fix: prevent totalTokens inflation in transcript persistence path (#125333) — Fixes #125333 What Problem This Solves #123065 added an "unavailable context" barrier that only applies when assistant.api === "cli". For every other provider the cumulative assistant usage was still...
- fix(ollama): reject invalid UTF-8 in streaming NDJSON responses — What Problem This Solves Ollama streaming responses were decoded with a non-fatal UTF-8 decoder, so corrupted bytes inside an NDJSON response silently became replacement characters. An incomplete...
- fix(codex): unify conversation permission policy — What Problem This Solves Split authorization ownership allowed conversation recovery, dynamic Gateway execution, and recovery compaction to consume raw or stale permission overlays after native Codex...
- fix: prevent stale transcript projections from publishing — Related #126914 Depends on #126943 Problem A transcript projection can be prepared from one source and publish after a same-length source replacement. Sequence numbers alone cannot distinguish those...
- fix(plugins): surface oversized-file warnings in doctor migration reads — What Problem This Solves Three plugin doctor-contract files read legacy state files during openclaw doctor migration with a TOCTOU-prone pattern: statSync(path) checks file size, then a separate...
- fix: keep subagent reconciliation case-sensitive — Fixes #122047 What Problem This Solves Fixes an issue where a missing subagent session could be reconciled as completed from a different sibling session whose key differed only by letter case. Why...
- feat(security): require acknowledgement for install policy warnings — New behavior An external security.installPolicy command can now return warn, letting an authorized operator review a suspicious plugin or skill install before deciding whether to continue. -...
- fix(onboarding): stop recommending tools setup cannot activate — What Problem This Solves Fixes an issue where the no-AI onboarding state told users to install Pi, OpenCode, Gemini CLI, Kimi Code, or Grok Build and then "check again", even though guided setup...
- feat: persist stateful transcript display semantics — Related: #126914 Depends on: #126939 Part 2 of the durable transcript display-row projection series. Problem Visible chat rows are stateful across transcript events. Message-tool calls and results,...
- perf(ci): split Telegram prerelease tests — What Problem This Solves Plugin Prerelease placed Telegram's entire extension suite inside one generic balanced shard. In the successful August 24 run...
- fix(ui): remove visible widget scrollbar tracks — Closes #128492 What Problem This Solves Fixes an issue where users viewing scrollable HTML widgets on the Control UI dashboard would see a full-height native scrollbar rail instead of the app's thin...
- build(deps): bump the android-deps group across 1 directory with 26 updates — Bumps the android-deps group with 26 updates in the /apps/android directory: | an
- fix(github-copilot): make the Copilot integration id configurable for *.ghe.com tenants — What Problem This Solves Since 2026.7.2-beta.7 (via #114282), every GitHub Copilot request sends Copilot-Integration-Id: copilot-developer-cli. That is the right default for github.com fine-grained...
- feat(update): assess version-bound companions before package updates — Related: #122019 What Problem This Solves Package updates can stop a working Gateway and replace OpenClaw before operators learn that a configured release-bound companion is missing or belongs to a...
- feat: allow global pairing reply templates — Closes #109240 What Problem This Solves Operators using DM pairing currently send a fixed OpenClaw-branded reply to every new requester. They cannot localize it, align it with an assistant identity,...
- fix: /models stops working after an unrelated config edit and never recovers — Related: #128515 AI-assisted: implemented by a Codex-based agent, reviewed and validated by Claude (Opus 5). What Problem This Solves Fixes an issue where operators who edit a config key that needs...
- fix(agents): surface failed subagent completion notices — What Problem This Solves Related to #128449. This PR fixes the silent direct-notice path only; the issue remains open to track its separate outcome='completed' telemetry misclassification. When a...
- feat: persist bounded transcript display rows — Related: #126914 Part 1 of the durable transcript display-row projection series. Follow-up layers are #126943 and #126947. Problem OpenClaw stores canonical transcript events, but raw events do not...
- OpenClaw repo stats for 2026-08-24 — 387284 stars, 81328 forks, 3135 contributors
- OpenClaw sponsors (92 total) — 92 sponsors supporting OpenClaw
- fix: show current NVIDIA evals from source-backed skills — Resolve current SkillEvaluator results through the GitHub source row when skills do not carry githubCurrentRepo, returning null instead of error/missing data when unavailable.
- docs: add avoid-ai-writing, model-hierarchy-skill, openclaw-search-skills
- Add skill: dkistenev/agent-inner-life
- Add skill: programmingwtf/repo-standardizer
- Add scvd-tab to Productivity & Tasks
- Update install instructions
- 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...
- Claude Code refuses requests or charges extra if your commits mention "OpenClaw"
- Tell HN: Anthropic no longer allowing Claude Code subscriptions to use OpenClaw
- Anthropic says OpenClaw-style Claude CLI usage is allowed again
- The Status Quo of AI in Software Development (2026) — Artificial Intelligence in 2026: From Companion to Infrastructure Artificial Intelligence...
- 6 Open-Source AI Automation Tools for Self-Hosted Business Workflows — Self-hosting is becoming more interesting as businesses start using AI for more than simple chat.Continue reading on Medium »
- Grok Bot vs OpenClaw: I Spent 40 Hours on This. OpenClaw Is Better Right Now — The honest answer on power, models, plugins, pricing, privacy, cloud IPs, LinkedIn risk, and who should still choose Grok Bot.Continue reading on Medium »
- nanochat — Complete LLM chatbot implementation, offering tokenization, pretraining, finetuning, evaluation, inference, and a simple web UI. Runs on a single 8xH100 node, features a hackable codebase,...