October 6, 2026
A session for every bug: Capy coordinates AI agents working in parallel
In Capy, you can assign multiple tasks to separate sessions and let Captain coordinate them.
Garry Tan
@garrytan
I'll admit, my work on gbrain is a bit chaotic. A bug or issue comes in, and I spin up a separate session to investigate. But the coordination between sessions in Capy and conflict prevention through a "virtual standup" during parallel work are state of the art. With autoplan and eli5 from GStack, I'm flying
· 35.4K views
Since Oct 1, Captain receives subordinate sessions' direct replies to the user, including approved merges of changes. Previously, it missed them.
Capy runs AI agents in the cloud to work on GitHub repositories. The documentation recommends a separate session for each task: you do not have to add a bug fix to a conversation about another improvement.
A task coordinator. Captain has been available as an experimental feature since Sep 24. It is a long-running session that manages other sessions as its team. Enable it through Settings → Experiments for yourself or your organization.
In version 0.4.2, a session added to the team while running comes under Captain Drive's control on its next step. Garry Tan calls the coordination of parallel work and conflict prevention through a "virtual standup" state of the art.
To get started, sign in at capy.ai, install the Capy GitHub App, select repositories and a model, then submit a task. In Settings → Models, you can connect your Codex or SuperGrok subscription or an API key.
A plan without the interrogation. In GStack, the `/autoplan` command replaces answers to 15–30 intermediate questions with six principles. The agent brings contentious decisions to you for final approval. Plan reviews run in this order:
1. CEO: a product-level check. 2. Design: interface architecture, if there is a UI. 3. DX: developer experience, if the changes are aimed at developers. 4. Eng: engineering review.
Install GStack for Claude Code with `git clone --single-branch --depth 1 https://github.com/garrytan/gstack.git ~/.claude/skills/gstack && cd ~/.claude/skills/gstack && ./setup`. Git and Bun are required; Bun 1.4.2 or later is recommended.
Install published skills in a Capy repository with `npx skills add <owner>/<repo> --skill <name> --agent universal`. Commit the result. Capy reads instructions from `.agents/skills/` and `.claude/skills/`.
As of Oct 6, Capy Lite costs $20 per month and includes $20 in usage credits. You can add unlimited members. For parallel runs, the plan provides 128 GB of total RAM, with each cloud machine limited to 4 CPU and 16 GB of RAM.
The first 7 days cost $1 with a verified card.
A cloud session keeps running after you close the tab and automatically receives CI, review, and merge events.
Source