Skip to main content

A CLI to get training data from your own coding agents. Track rollouts and codebase snapshots at every turn. The data can be used for SFT, RL, and Continual Learning.

Project description

Rollouts

A CLI for Continual Learning with your own coding agent sessions. Track rollouts and codebase snapshots at every turn. The data can be used for SFT, RL, and Continual Learning. Currently supports OpenCode and uses Prime Intellect for hosted RL training.

Quick Start

# Install rollouts
uv tool install agent-rollouts

# Install the OpenCode Rollouts plugin.
# If GitHub repo defaults are not configured yet, setup will prompt for them.
rollouts setup

# Start a learn session
rollouts learn start my-session --config /path/to/rl.toml

# Inspect learn sessions
rollouts learn list
rollouts learn status my-session

# Restart the latest run in the same session if it failed or was stopped
rollouts learn restart my-session

# Or override the config path when restarting
rollouts learn restart my-session --config /path/to/rl.toml

Make sure prime, opencode, git, gh, and hf are installed first.

If you plan to use rollouts learn, make sure your Prime environment is also configured.

Installation

Using uv

Install from PyPI:

uv tool install agent-rollouts

Upgrade later:

uv tool upgrade agent-rollouts

Usage

Learn Sessions

  • rollouts setup installs the OpenCode Rollouts plugin and prompts for GitHub repo defaults if they are not configured yet.
  • rollouts remote defaults set is required before rollouts learn start because learn sessions push tracked snapshots and may auto-create GitHub repos.
  • rollouts learn start <session> --config <path> creates the learn session, stores the config, syncs the dataset, creates run #1, and starts the initial Prime run.
  • rollouts learn list shows your learn sessions.
  • rollouts learn status <session> shows the latest run, current Prime state, config path, and restart lineage.
  • rollouts learn restart <session> restarts the latest failed or manually stopped run in the same session after confirmation.
  • rollouts learn restart <session> --config <path> overrides the stored config path for the restarted run.

If an older run was created before config-path tracking existed, rollouts learn restart may require --config the first time you retry it.

Local Repo Setup

For local development:

uv sync --extra dev

To install the current checkout as a tool:

uv tool install --editable .

You usually do not need to pass --python here as long as uv selects a compatible interpreter for the project.

Development

Quality checks:

uv run ruff check .
uv run ty check src

Pre-commit hooks are configured for:

  • ruff --fix
  • ruff format
  • ty check src/

Install the Git hook locally with:

uv run pre-commit install

Run hooks manually with:

uv run pre-commit run --all-files

Project details


Download files

Download the file for your platform. If you're not sure which to choose, learn more about installing packages.

Source Distribution

agent_rollouts-0.1.3.tar.gz (129.3 kB view details)

Uploaded Source

Built Distribution

If you're not sure about the file name format, learn more about wheel file names.

agent_rollouts-0.1.3-py3-none-any.whl (39.2 kB view details)

Uploaded Python 3

File details

Details for the file agent_rollouts-0.1.3.tar.gz.

File metadata

  • Download URL: agent_rollouts-0.1.3.tar.gz
  • Upload date:
  • Size: 129.3 kB
  • Tags: Source
  • Uploaded using Trusted Publishing? No
  • Uploaded via: uv/0.9.7

File hashes

Hashes for agent_rollouts-0.1.3.tar.gz
Algorithm Hash digest
SHA256 0582615e8707e001fe53250472335bfacd5091623586c29943405d1fc6e43d34
MD5 9d7a82e42b4633a81891456e29fe6a24
BLAKE2b-256 192621bd62780a0077ed21f037a69b801318c3514e67bb2613744cdd5400fbb3

See more details on using hashes here.

File details

Details for the file agent_rollouts-0.1.3-py3-none-any.whl.

File metadata

File hashes

Hashes for agent_rollouts-0.1.3-py3-none-any.whl
Algorithm Hash digest
SHA256 6c6044224221358052bcd5af785e2184142706c51555c38c9ab61b12ed20cab4
MD5 c81841ce8c6008ff648aafe487892669
BLAKE2b-256 eae556acf9ce2beb792b514425f11e9525d67ee84e44bc8d1c31c2c987aafb8a

See more details on using hashes here.

Supported by

AWS Cloud computing and Security Sponsor Datadog Monitoring Depot Continuous Integration Fastly CDN Google Download Analytics Pingdom Monitoring Sentry Error logging StatusPage Status page