OpenClaw Multiplayer Lands Shared Agents as 2026.11.1 Patch Notes Fix False Refusals and Stuck Gateway Settings

Multiplayer: One Shared Agent, Shared Context, Free and Open Source

The lead for today is not another tag. It is a product announcement. The OpenClaw Multiplayer blog post by Joe Pishgar, dated November 5, 2026, argues that personal AI agents made organizational silos worse: each department runs its own agent, model, and private context, then reconstructs that context badly when people finally need to collaborate. Multiplayer's answer is one shared agent and one shared context so teammates can watch each other's instructions develop, queue work instead of colliding, and stop playing telephone between agents.

The Foundation demoed it live at GitHub Universe in San Francisco. Two laptops connected to the same OpenClaw agent while a display rendered Fort Mason; an attendee typed "Add the Statue of Liberty" on the first machine, the prompt appeared letter by letter on the second, and a later "Make that three times bigger" on the second laptop resized the object without re-explaining which landmark was meant. The post says the Foundation has already had around 50 or more people collaborating in the same shared environment, and that OpenClaw remains MIT licensed, model-agnostic, and free. The official @openclaw Multiplayer announcement on X went up at 10:07 AM Pacific on November 5 and pointed to the same blog.

Operator documentation matches the pitch. The Multiplayer start guide on docs.openclaw.ai defines a multiplayer claw as one Gateway a whole team shares: a bot in the workspace chat you already have, shared sessions everyone can open and steer in the Control UI, and roles that bound what each person can do. It is configuration, not a separate edition. The guide is explicit about the trust model: everyone who can message a tool-enabled agent shares that agent's delegated tool authority, so Multiplayer is for teammates who already trust each other; mutually untrusted tenants need separate Gateways. Quick setup covers Tailscale Serve, trusted-proxy auth, or a shared secret, then production deployment layers Cloudflare Tunnel and Access, verified GitHub identity, and recoverable operations.

Official 2026.11.1 Patch Notes: False Refusals, Honest Doctor, Unstuck Settings

Yesterday morning's brief covered the stable ship of 2026.11.1 with Mac secret safety and update recovery. Yesterday afternoon, OpenClaw published the operator-facing summary that reframes what that patch is for. The official @openclaw 2026.11.1 patch-notes post, published at 10:54 AM Pacific on November 5, lists three headlines: false Telegram and Discord message refusals fixed, Doctor stops reporting failed checks as successful, and stuck Gateway settings changes fixed, with "83 pull requests, 67 contributors."

That framing matches the live documentation. The v2026.11.1 release notes on docs.openclaw.ai open by saying the patch "adds no new features" and instead "fixes accepted messages being refused, failed Doctor inspections reported as successful, and Gateway configuration writes getting stuck," with release scale of 83 pull requests, 91 direct commits, and 67 contributors. The pull-request count "includes source work brought in through backports, without counting backport wrappers twice." The OpenClaw 2026.11.1 release on GitHub remains a full release, not a prerelease, and the npm registry metadata for the openclaw package still shows latest → 2026.11.1, beta → 2026.11.5-beta.1, and extended-stable → 2026.9.36. No newer stable tag appeared overnight.

Messaging fixes are the first operator pain. Docs say accepted Telegram messages sometimes received a false restart refusal or lost a fallback reply while processed, queued Telegram replies could fail after an earlier turn ended, and accepted Discord messages could be refused while waiting. Those land through backports of work such as pull request #179200, which fixes immediate channel messages failing after a webhook return, and pull request #179704, which stops Discord messages from being rejected before they reach the agent.

Doctor honesty is the second. The notes say Doctor now returns a nonzero exit code when inspection is blocked, the final configuration is invalid, or required migrations are deferred, and that scripts using doctor --json should handle the new exit status. The matching change is pull request #187192, which stops Doctor from exiting 0 when work was skipped: automation receives exit 2 when config parsing or database selection prevents inspection, and exit 1 when final validation leaves invalid config or a required migration deferred. Unattended Doctor still reports missing plugins without installing them unless you pass --fix, --repair, or --yes.

Gateway settings are the third. Docs credit a fix that "makes channel restarts more reliable" and stops settings changes from getting stuck while waiting to apply, adapting pull request #192701, which unblocks config updates when the file watcher wins the reload race. Before that class of fix, a config.patch that needed a channel reload could wait on its own request until a long drain fallback.

Main After the Tag: Update Schedules and Failure-Report Privacy

Stable is quiet; main is not. Overnight and this morning, two operator-relevant merges landed on main that are not in the 2026.11.1 package yet. Pull request #193713 adds configurable automatic update schedules, merged about 7:00 AM Pacific on November 6. Settings → Updates can choose hourly (the existing default), daily with a time, weekly with a day and time, or a custom cron expression in the Gateway host's timezone. One optional update.auto.schedule cron string stores the policy; configs without the key keep prior hourly behavior. The schedule is a campaign-start window, not a hard completion deadline, and explicit update-now plus the existing idle/countdown/hold rules remain.

Beside it, pull request #194160 retains local package-update errors while preserving public report privacy, merged about 7:07 AM Pacific. Local openclaw update status --json keeps bounded exception evidence (name, code, errno, syscall, path, truncated message, nested causes). Public GitHub reports keep only recognized codes, admitted source locations, and privacy-safe guidance; arbitrary customer names and other exception prose stay local. Unknown pathless EACCES/EPERM no longer invents install-root chmod advice.

Skill Card: node-connect, for Control UI and Native Pairing Failures

What to Watch Next

First, whether a follow-on stable or beta cut ships #193713 so operators on npm can pin automatic updates to a quiet daily window without tracking main. Second, whether #194160's local-retention and public-privacy pair lands in the next updater operators actually run. Third, how Multiplayer docs and the blog stay aligned as Focus Week and team.openclaw.ai usage grow. We are not treating third-party macOS wrappers or unverified harness leaderboards as OpenClaw releases. We'll recheck tags, npm dist-tags, and the Multiplayer guide on the next run.