OpenClaw 6.5: The 377K-Star Open-Source AI Assistant Fixes Pain Points
OpenClaw shipped v2026.6.5 stable. With 377K stars, it is the open-source personal AI assistant that plugs into nearly every IM platform. This release fixes three real-world failures in long sessions and IM channels.
💡 What You Will Learn
OpenClaw shipped v2026.6.5 stable. With 377K stars, it is the open-source personal AI assistant that plugs into nearly every IM platform. This release fixes three real-world failures in long sessions
📜 Table of Contents
OpenClaw 6.5: the 377K-star open-source assistant fixes more sharp edges
OpenClaw released v2026.6.5 stable. At 377K stars, it positions itself as your own personal AI assistant, cross-platform and cross-channel, connecting to WeChat, Discord, WhatsApp, QQ and Feishu.
No headline features this time - three hard bugs. That is worth attention: the project has moved from feature racing to long-run reliability, which is the line between a demo and something you leave running.
Three fixes and what each solves
1. Anthropic long-session crash recovery. Past a certain length, signature validation fails and the session dies; you used to resend manually. Now a validation failure triggers an automatic retry. The root cause is server-side session signing: the longer the context, the more likely signature and cache state diverge. Retrying is not a cure, but it converts user intervention into something the system absorbs.
2. QQ bot no longer leaks thinking tags. Reasoning sits inside <thinking> tags that should be stripped before sending. The QQ channel skipped that step and users saw the model thinking aloud.
3. Feishu 429 retries cover all three codes. Feishu rate limiting surfaces as HTTP 429, business code 230020 and business code 11232; retries previously missed some. Rate limiting is the number one failure mode on IM channels - busy groups mean dense API calls and inevitable 429s. Backoff retries decide whether a bot drops messages or delivers reliably.
Two additions
Parallel search plugin is built in. Search used to be a separate install; web access is table stakes for a personal assistant.
ClawHub installs skills straight from GitHub repos. The ecosystem opens beyond the official marketplace, at the cost of judging safety yourself - read the source of third-party skills, especially any that need tokens.
Upgrading
npm i -g openclaw@2026.6.5
Run the self-check first:
openclaw doctor
It checks config, dependencies and channel connectivity. Doctor before upgrade, so you know which problems predate the change.
Teams and anyone leaning on Feishu or QQ should upgrade soon; those are the channels this release targets. Personal users can wait a day for community feedback.
Rollback plan
- Back up the config directory
- Note the current version so you can
npm i -g openclaw@<old> - Verify on a secondary channel (a Telegram DM is enough) before touching production groups
The trade-off
Upside: broad channel coverage, self-hosted, your data stays local, fast release cadence.
Cost: every channel is an adapter, and adapters break; IM platforms change APIs and anti-abuse rules constantly. Choosing this project means accepting that you operate it.
FAQ
Does upgrading lose config? Normally not, but a backup is free insurance, especially across major versions.
Feishu still silent? Check whether 429s persist in the log, then verify app permissions and callback URLs. Retries absorb bursts, not misconfiguration.
Can I use one channel only? Yes, and you should start there: prove message flow and tool calls before adding a second.
Are GitHub-installed skills safe? Depends on the repo. Before installing, look for outbound network calls, local file reads and API key requirements. If unsure, skip it.
How does it compare to a hosted assistant? Maintenance is your job. In exchange, data and channels stay under your control and local models are an option.
Related
Written by our editorial team; tools listed here are tested or verified against public sources. Links point to official sites or GitHub repos for reference only — no paid placements.