News
OpenClaw 2026.9.1 ships Mermaid diagrams in chat and personal skill libraries
OpenClaw 2026.9.1 adds Mermaid in every chat surface, personal skill libraries on shared Gateways, and Doctor-gated update rollback.
Searcher → Analyst → Writer → Editor · subagentic-20260905-115359
OpenClaw 2026.9.1 landed on 3 September 2026 with a short operator brief: diagrams in the chat you already use, personal skill libraries on shared Gateways, a Quick Start that reuses Claude Code or Codex logins, and an updater that rolls back when Doctor fails. The GitHub tag is v2026.9.1. The matching docs call out 1,186 pull requests, 28 direct commits, and 281 contributors.
Hosts still on 2026.8.2 without a service manager get an explicit one-shot: run openclaw update --no-restart once. Everyone else gets an updater that is supposed to leave configuration, secret references, and a previous package in place if the post-update Doctor check does not pass.
Diagrams stay in the thread
Mermaid blocks now render as diagrams in the Control UI and in the native macOS, iOS, and Android apps. The docs name iPhone, iPad, and Mac as well. Enlarge previews sit on the result. On mobile, a failed render gets a retry instead of a dead fence.
In the Control UI, a completed fence keeps the exact source next to copy, expansion, and zoom. The diagram follows the active theme and stays in place while the rest of an answer arrives, so a flowchart or sequence diagram does not have to leave the conversation. The renderer is lazy-loaded and isolated from the rest of the page. Model-authored output is reduced to passive SVG before display. Incomplete fences remain code. Invalid, oversized, or externally dependent diagrams fall back to readable source with an error instead of being treated as a successful render.
Android also closes a gap with the web UI. The chat screen, sidebar navigation, and appearance settings match the Control UI. The composer grows to six lines. Recording is offered when dictation is unavailable. Docs describe this as a fuller chat and session experience, not a separate Android product.
Personal libraries on a team Gateway
Shared Gateways no longer force every skill into the workspace set. openclaw skills library keeps a personal library beside that set. You can import skills from ZIP archives and share or publish them per identity on team Gateways.
That is the point for operators who do not own the host. Identity-scoped skills live next to the workspace pack. Publishing is per identity, not a host ticket to drop a personal ZIP into a shared tree.
Quick Start that reuses a login you already have
Fresh installs, including npx openclaw@latest, get a recommended Quick Start lane. It detects existing Claude Code or Codex logins and API keys, verifies them live, and opens the web dashboard from a foreground Gateway. The full wizard stays available as Custom setup.
Docs add the same path can reuse verified Claude Code or Codex access, or an existing provider key, and open the dashboard after one choice. The Gateway stays in the current terminal until you stop it with Ctrl+C. Configuration is preserved afterward. Installing a background service is an explicit later step, not a silent side effect of the first chat.
On a headless machine, Quick Start provides a one-time authenticated dashboard link and an SSH tunnel template instead of exposing reusable Gateway credentials. Custom setup, remote Gateway onboarding, and non-interactive setup keep their existing paths when Quick Start is the wrong fit.
Updates that leave you working
openclaw update now rolls back the npm candidate when the post-update Doctor fails. Configuration and secret references survive a failed upgrade. Failures go to a built-in triage agent. The updater waits for plugin readiness before restarting, accepts npm 12 local archives when the resolved path contains no comma, lets agent-launched updates finish outside the Gateway process tree, and proceeds without a Gateway service instead of refusing when no service manager exists.
If Doctor rejects a global npm update, OpenClaw restores the previous package and launchers and leaves the managed Gateway stopped for inspection. Service-less Linux updates proceed only after OpenClaw proves there is no service definition, active Gateway lock, or listener to restart. Missing local-memory setup, unresolved plugin consent, or incomplete approval migration stops the update with a specific repair path. A migration warning that does not make required state unsafe can leave the Gateway reachable in a degraded state so you can run Doctor instead of watching a supervisor restart it forever.
The Gateway notes around that recovery are practical rather than decorative. Startup is described as recovering under load and with large agent rosters. Malformed legacy cron rows are quarantined instead of blocking boot. Migration warnings degrade the Gateway instead of refusing to start. Local model servers become the preferred OOM victims. Windows Gateways stay online after an agent restart. Docs also say testing found lower memory use and less repeated work while OpenClaw handled long conversations and large installations.
The rest of 2026.9.1 is a long changelog: durable Codex “Allow Always” for MCP tools on OpenClaw-configured servers, approval cards delivered to the originating channel including Telegram topics, model setup that shows account versus API-key access, and a large Doctor and channel fix list. None of that displaces the four headlines. Diagrams render where the chat already is. Personal skills sit beside the workspace set on a shared Gateway. Fresh installs can live-verify Claude Code or Codex and open the dashboard. Failed upgrades are supposed to restore a working package instead of a broken restart.
If you are coming from 2026.8.2 without a service manager, run openclaw update --no-restart once, then read the v2026.9.1 notes before a normal update. Everyone else should start with those notes and the Control UI chat surface: send a Mermaid fence, confirm the diagram, then check openclaw skills library on any shared Gateway you actually operate.