Skip to content
Back to use cases
AI agents

Run your AI agents on infrastructure you can leave

Your agent's instructions stay in your git repo. It runs on your own Anthropic or OpenAI key. The runner that executes it is MIT-licensed open source that you can build and run anywhere Docker runs.

What runs where

Every part has its own way out.

  • Part: Agent definition (CLAUDE.md, skills, prompt)

    Where it lives: Your git repository

    Stays where it is. Nothing OSC-specific is added to it.

  • Part: Runner

    Where it lives: Claude Runner or Codex Runner

    MIT-licensed. Build the same image with docker build from the public repo.

  • Part: Model

    Where it lives: Your Anthropic or OpenAI account

    Billed to that account, not to your OSC tokens.

  • Part: State

    Where it lives: PostgreSQL and MinIO

    Unmodified upstream open source. Take your data with pg_dump and any S3 copy tool.

  • Part: Schedule

    Where it lives: My Agent Tasks

    A feature of the OSC app, on Personal and up. It holds prompts and schedules.

Three steps

1

Point it at your repo

Set SourceUrl to the git repository that holds your agent's CLAUDE.md, skills or code. Add GitToken if the repo is private.

2

Bring your own key

Claude Runner takes AnthropicApiKey or ClaudeCodeOauthToken. Codex Runner takes CodexApiKey or OpenaiApiKey.

3

Give it a job

Set Prompt. Run it once, or put it on a schedule in My Agent Tasks. Add OscAccessToken if the agent should reach OSC services over MCP.

The two runners

Both are job services in the catalog. They clone your repo, run the prompt and stop.

Claude Runner

birme-claude-runner

Runs Claude Code headless against your repo. Auth: AnthropicApiKey or ClaudeCodeOauthToken.

Open in the catalogSource code (MIT)

Codex Runner

birme-codex-runner

Runs Codex against your repo. Auth: CodexApiKey or OpenaiApiKey.

Open in the catalogSource code (MIT)

What does not happen when you leave

If you stop using OSC, this is the list of things that stay as they are.

  • Your repo is unchanged. The agent tasks themselves are just prompts and schedules.
  • Your model account is unchanged. Usage was billed there all along.
  • The runner still works. Each repo README documents docker build and docker run with the same SOURCE_URL.
  • Your data comes with you via pg_dump and any S3 copy tool.

One worked example

A scheduled agent that tags a media library, with its own recipe.

See the media library agentMachine-readable recipe (JSON)

Other agents you can run here

Questions

Who pays for the model when an agent task runs?

Agent tasks run Claude or Codex using credentials you configure in the Credentials panel. The AI model usage is billed directly against your Anthropic or OpenAI account, not your OSC token balance. OSC tokens are consumed only by the infrastructure services your tasks operate on.

Which plan do I need to schedule tasks?

Scheduled agent tasks are available on Personal and up. Personal is 15 EUR per month excl. VAT. The free Basic tier includes the full service catalog and MCP access.

Is the scheduler open source?

No. My Agent Tasks is a feature of the OSC app. The two runners are MIT-licensed open source, and the services your agents use are unmodified upstream open source. The agent CLIs inside the runners come from their vendors and are not part of that claim.

How do I leave?

The agent tasks themselves are just prompts and schedules. If you stop using OSC, your repo is unchanged, your data is portable via standard database tools, and there is nothing OSC-specific inside your codebase. Data leaves with pg_dump and any S3 copy tool, and the runner image builds from its public repo.

Run your first agent

Create a free account, add your model key and point a runner at your repo.

Start freeBrowse the catalog