Open source / Python / Local state

Run the agent.
Verify the result.

Mobius repeats one coding task until the agent reports completion and your project checks pass.

pipx install mobius-agent
View source
Built-in adapters CodexClaude CodeGemini CLIAny command

Example

One objective.
One completion rule.

Mobius keeps the objective unchanged between turns. A run completes only when the agent signal and every configured check pass.

Illustrated Mobius terminal run completing after tests and lint pass
Example runNo telemetryJSON state

How it works

A clear completion rule.

Mobius runs outside the agent. It sends the same objective on each turn, runs your checks, and records the result.

01

Set one objective

Describe the result you need. Mobius keeps that objective unchanged for the full run.

mobius run "Fix the checkout tests"
02

Run your agent

Use a built-in adapter or configure another command. Iteration and process limits apply to every run.

--agent claude
03

Run project checks

The agent signal and every validation command must pass. Failed output is included in the next turn.

--check "python -m pytest"

Run control

Set limits.
Inspect the state.

Mobius runs in the foreground. Each run has an iteration limit, process timeout, and project-local state file.

  • Iteration and timeout limits
  • Project-local JSON state in .mobius/
  • Multiple validation commands
  • Prompt preview with --dry-run
01

Foreground execution

Output remains visible and the run stops when the process exits.

02

Project validation

Use the test, lint, type-check, or build commands your project already trusts.

03

Readable state

JSON state stays in .mobius/ and can be inspected, resumed, or removed.

Quick start

Install and run.

1

Install with pipxpipx install mobius-agent

2

Detect your project checksmobius init

3

Start the loopmobius run "your objective"

Start with one objective.

Review the configuration, preview the prompt, and start the run.

Read the documentation