Rush 0.1.0-alpha.50 — Bring any coding agent to the cloud

Until now, a cloud run meant Claude. In this build you can connect a Grok, Cursor, GitHub Copilot, or Codex key once, and those agents run in the cloud exactly like Claude does — no repo required, the same ⌘+Enter, the same live detail view. Rush hands your key to the vendor CLI securely, runs it in an isolated cloud sandbox, and streams every step back to you.

When the run finishes, its full transcript and token-and-cost accounting come back with it, rendered in the same detail view you already use for local runs. You pick the agent for the job — one model for refactors, another for reviews — instead of being locked to a single one.

Whether an agent runs on your machine or in the cloud, it opens the same detail view: the activity timeline on the left, every tool call and message on the right as it happens. You see exactly what the agent did — the files it read, the commands it ran, the code it wrote — instead of a summary after the fact.

Cloud transcripts now keep every assistant and tool turn end to end, so nothing vanishes inside a cloud pod, and a run no longer fails with a 502 just because a connected account couldn't refresh a token — it degrades gracefully and tells you what happened.

Connected-account tokens for cloud dispatch are now vault-backed per user, and connecting Google or Microsoft is more robust. Bring any MCP server as a connector — paste a URL, pick the transport, flag OAuth if it needs it — and reuse a saved connection across every agent, choosing per agent which ones it can touch.

Rush also tells you up front when an agent needs an account you haven't connected yet, and remembers when you dismiss a prompt so it stops re-asking. Disconnecting Google or Microsoft now removes exactly that account's cloud credentials — no stray tokens left behind.

Schedule any agent — Claude or one of the new ones — to run on a recurring cadence and then forget about it. Runs fire unattended, permissions accepted automatically so nothing stalls waiting on a click, and anything due while Rush was closed is caught up the next time you open the app.

New this build: finish setup straight from the menu bar, with a Settings → Setup checklist that shows exactly what's left to connect.

• Sign-in stays put. No more looking logged in against a dead session, no login loop that wiped the session you already had, and "Welcome back / Continue" never clears a valid session. You log in once and stay in.
• Your final report, front and center. Multi-artifact research sessions (Rabbit Hole, Grounded Scientist) now pin and auto-open the finished report instead of burying it under every intermediate scratch file.
• The full timeline, live. Sub-agent steps and inline renders (charts, image search) now appear in the session timeline as they happen, not only after the fact in a separate tab.
• Player artifacts play inline. Meditations, audio and video render and play all the way to the end right in the session view, instead of stopping a few seconds in.
• The floating agent card keeps up. The cockpit that hovers over your browser during a run advances with each step, instead of freezing on the first activity.
• Clearer errors. When an external agent's key or auth fails, you get a plain-language message — not a raw error blob leaking onto the screen.
• Self-serve developer access. rush dev enable unlocks publishing instantly for eligible accounts, no manual review round-trip.