v0.122.9

## Highlights

- **Embed an agent on your website.** You can now put an agent on any web page as a chat window in an iframe. Visitors chat with the agent without an Agenta account. See the embed guide in the docs.
- **Use your Claude subscription.** You can connect a Claude Pro or Max plan to a project by pasting a token from `claude setup-token`. Agents that use the Claude Code harness then run on that plan instead of an API key.
- **Steadier agent runs.** The runner no longer ends a turn while it waits for the model's first response. Paused in-process conversations no longer fill the memory and block new sessions. When the runner refuses a turn, the API now returns 409, 503 or 422 instead of 500.
- **Sign-in fixes.** When a one-time code is the only way to sign in, the code is sent at once. A failed security check now returns 401 with a clear message instead of 500.
- **Organization limit.** Self-hosted deployments can limit how many organizations one user owns with `AGENTA_ACCESS_MAX_OWNED_ORGANIZATIONS`. If you don't set it, there is no limit.
- **New tutorials.** The docs have a Tutorials section with three video tutorials.

## What's Changed

* [feat] Embed an agent on any website as a chat iframe by @mmabrouk in https://github.com/Agenta-AI/agenta/pull/7662
* feat: connect a Claude subscription with a pasted setup-token by @mmabrouk in https://github.com/Agenta-AI/agenta/pull/7691
* fix(sdk): Return 4xx when agent invoke refuses the caller's configuration by @mmabrouk in https://github.com/Agenta-AI/agenta/pull/7678
* fix(runner,sdk): Report agent turn admission refusals as 409, 503 or 422 instead of 500 by @mmabrouk in https://github.com/Agenta-AI/agenta/pull/7679
* [fix] Stop extra organizations from collecting free credits by @mmabrouk in https://github.com/Agenta-AI/agenta/pull/7681
* docs: Tutorials section with three video tutorials by @mmabrouk in https://github.com/Agenta-AI/agenta/pull/7677
* [fix] Return 401 instead of 500 when Turnstile rejects a sign-in by @mmabrouk in https://github.com/Agenta-AI/agenta/pull/7689
* fix(frontend): keep subscription names readable in the model picker by @mmabrouk in https://github.com/Agenta-AI/agenta/pull/7686
* fix(runner): stop the turn watchdog from ending a turn while the model's first response is pending by @mmabrouk in https://github.com/Agenta-AI/agenta/pull/7688
* fix(frontend): show where a subscription comes from only when the plan appears twice by @mmabrouk in https://github.com/Agenta-AI/agenta/pull/7692
* fix(runner): stop parked in-process conversations from filling the heap and blocking new sessions by @mmabrouk in https://github.com/Agenta-AI/agenta/pull/7682
* fix(frontend,docs): wrap the model picker caption and list the run limit error codes by @mmabrouk in https://github.com/Agenta-AI/agenta/pull/7694
* fix(hosting,docs): give web-mobile the internal API URL and correct embed and error docs by @mmabrouk in https://github.com/Agenta-AI/agenta/pull/7695
* [fix] Send the sign-in code at once when a code is the only way in (from #7659) by @mmabrouk in https://github.com/Agenta-AI/agenta/pull/7673
* fix(api): title channel and embed sessions with the person's words, not the attribution line by @mmabrouk in https://github.com/Agenta-AI/agenta/pull/7696
* fix(hosting): map host.docker.internal for the runner; count codex MCP tool names in the gate by @mmabrouk in https://github.com/Agenta-AI/agenta/pull/7697

**Full Changelog**: https://github.com/Agenta-AI/agenta/compare/v0.122.8...v0.122.9