freesolo-agent 0.1.4

Thin CLI for queuing Freesolo backend training jobs.

Freesolo Agent CLI

Thin interactive command-line client for queuing Freesolo backend training jobs.

The public package does not contain the Codex/GitHub/training worker. It only:

• prompts for setup input
• verifies a Freesolo API key
• uploads a local dataset file or folder when one is provided
• enqueues /api/training/jobs
• polls /api/training/jobs/{job_id} until the backend worker finishes

The private worker implementation lives under backend/src/training_agent.

Install

pip install freesolo-agent

This installs the freesolo command.

From this repo checkout:

cd agent
uv sync
uv run freesolo setup

Setup

Run:

freesolo setup

or, from a checkout:

uv run freesolo setup

The CLI prompts for:

• Freesolo API key
• Freesolo org ID
• operation: draft, edit, optimize, or training
• source repository URL for draft
• target training repository URL for edit, optimize, and training
• optional branch
• optional dataset file or folder path
• optional backend-worker environment variables

For private repositories, install the Freesolo GitHub App when prompted:

https://github.com/apps/freesolo-agent/installations/new

The CLI prints progress while the backend job is queued and running. When the job finishes, it writes a compact JSON result containing ok, result, and jobId.

Operations

draft creates the initial Freesolo training contract from a source repo. It returns a targetRepoUrl for the generated training repo.

edit revises an existing training contract in the target training repo.

optimize generates and checks repo-level training files in the target training repo.

training runs the long training workflow from the current target training repo state.

Environment

The CLI does not require environment variables.

• Interactive setup prompts for the Freesolo API key.
• Dataset paths are uploaded before the backend job starts.
• Backend URL and job polling interval use CLI defaults.
• Private GitHub app credentials, Codex settings, Tinker dependencies, and training runtime configuration belong to the backend worker, not this CLI package.