Grain vs OpenAgents

Quick answer

OpenAgents is where agents and people coordinate. It pairs the coding agents you already run with a shared workspace of threads, files, a knowledge base, a task board and a browser everyone can watch, and one thread can route work between Claude Code and Codex. Grain is where finished work lives. Your agent writes files, Grain publishes them at a URL that stays up after the session, and an app can add Postgres, logins and scheduled jobs. Use OpenAgents to coordinate the work and Grain to publish the result.

Fit first

What is different, verifiably

Where OpenAgents is better

• A shared live browser. Agents research, click and fill forms while you watch, and you can take over at any time.
• Task routing between agents. You @mention Claude Code to review and Codex to write tests in the same thread.
• Agent setup. agn install claude-code installs, updates and configures a dozen agent CLIs.
• Team tools in one place: a task board, workflows, routines and an inbox for agent notifications.
• Full self-hosting under plain Apache 2.0, with code staying on your machine.

Where Grain is better

• A URL for the result. Pages, reports and apps publish with public, password or invite-only links. OpenAgents shares read-only snapshots of chat.
• Links that last. Published artifacts stay published after the agent session and after a user stops paying.
• A running app behind the URL, with Postgres, logins and scheduled jobs.
• Every artifact is plain files. grain pull fetches the whole workspace at once, with no per-file downloads.

Moving between them

1. From OpenAgents to Grain: download the shared files you need from the Files view. Work your local agents wrote is already on your disk.
2. Run grain init in the project folder so Claude Code, Codex and Cursor get the MCP config for Grain. Agents paired with OpenAgents keep working there.
3. Ask an agent to turn the files and knowledge notes into a page, report or app. Use the workspace's Postgres database for data that people update.
4. Run grain push, grain checkpoint -m "publish from OpenAgents" and grain share --public, or --password <pw> for a private link. Threads, the task board, routines and agent pairings stay in OpenAgents.
5. From Grain to OpenAgents: run grain pull and drop the folder into a workspace's shared files. Grain's database, logins, scheduled jobs, share links and custom domains do not carry over.

Choice rule

Choose OpenAgents

if you want your team and several coding agents in one workspace, handing tasks back and forth, with a shared browser and full self-hosting.

Choose Grain

if the output needs a lasting URL for people outside the workspace, with password or invite-only access, a database or scheduled jobs.

Use both

if agents coordinate in OpenAgents and one of them publishes the finished page or app to Grain with grain push and grain share.

Frequently asked questions