Grain with Claude Squad and Vibe Kanban

The relationship in one line

An orchestrator expands the amount of agent work you can run; Grain narrows that work into artifacts people can find, open and improve. Claude Squad is a local project for managing multiple AI terminal agents and their workspaces. Vibe Kanban supports a range of coding agents and, after Bloop's April 2026 shutdown announcement, continues as a community-maintained local project without its former remote collaboration services. Grain is neither a queue nor a multi-agent controller. It is the publishing destination that remains stable even when individual branches, terminals and tasks disappear.

What each layer owns

When you need both

Use both when parallel agent work produces something that people outside the agent console need. A product lead might request a launch-readiness workspace. One agent builds a dependency report, another assembles a risk dashboard and a third writes a customer-facing walkthrough. Claude Squad or Vibe Kanban helps the operator divide and monitor those tasks. Grain holds the combined result: a page linking the report and walkthrough, an app backed by Postgres for risk decisions, login access for the team and a scheduled refresh. The orchestrator can discard completed terminals and branches after integration. The Grain URL stays the shared reference, and comments on it become concrete input for the next agent pass.

When you need only one

Choose an orchestrator alone when the work ends as a code change in an existing repository. If developers review pull requests, merge them and rely on the application's existing deployment pipeline, a second artifact destination may add no value. Choose Grain alone when one coding agent is enough but its output needs a lasting URL. A report, prototype or internal app does not require multi-agent coordination merely because it needs publishing. Grain cannot decide task dependencies, isolate conflicting branches or arbitrate which agent result should win. Conversely, an orchestrator does not by itself provide Grain's sharing modes, app database, login layer, scheduled jobs or artifact comments. Add only the missing layer.

How it fits together

1. Define one deliverable. Write a brief such as: Create a launch-readiness workspace with a dependency report, a risk dashboard and a customer walkthrough. Add a Postgres table for risk owners, team login and a daily data refresh. A single outcome keeps parallel agents from producing disconnected demos.
2. Split the work in the orchestrator. Assign the report adapter, dashboard interface and walkthrough content to separate agent tasks. Claude Squad can manage multiple terminal agents; Vibe Kanban can coordinate supported local agents. Keep merge and conflict decisions in this layer.
3. Create the destination folder. In the integration branch or folder, run grain init. Grain creates the workspace and writes MCP configuration in .mcp.json, .codex/config.toml and .cursor/mcp.json.
4. Integrate before publishing. Have the coordinating agent bring the accepted files into the initialized folder, connect navigation and verify that the page, Postgres data path, login requirement and schedule agree with the brief. Grain does not choose among competing branches.
5. Push a coherent artifact. Run grain push, then grain checkpoint -m "launch readiness review". The checkpoint marks the assembled state, not every intermediate result created by each agent.
6. Share with the intended group. Run grain share --password <pw> or select invite-only sharing for internal review. Use grain share --public only when the artifact is meant for anyone with the link.
7. Route feedback back to work. Reviewers comment on the artifact. The agent can read those comments. Turn each accepted change into an orchestrated task, integrate the result, then push and checkpoint again.

Frequently asked questions