# facebookresearch/Pearl

A Production-ready Reinforcement Learning AI Agent Library brought by the Applied Reinforcement Learning team at Meta.

Repository: https://github.com/facebookresearch/Pearl
Canonical: https://ross.abutalabs.com/products/pearl
Language: Jupyter Notebook
License: MIT
License Family: permissive
Last push: 2026-08-19T01:48:27+00:00

## Health v2 (maintenance only)
Score: 74/100 (v2, computed 2026-09-03T02:20:16.233290+00:00)
- activity 98, release rhythm 35, longevity 88
- inputs: {"age_days": 1244, "days_push": 15, "days_rel": null, "gap_med": null, "n_releases_24m": 0}
- flags: no_releases
- formula: round(0.45*activity + 0.35*rhythm + 0.20*longevity); archived -> min(score, 10)

## Adoption (not part of the score)
Stars 3024, forks 205 (observed 2026-08-28T04:07:38.576188+00:00)

## What it is
Pearl is a production-ready reinforcement learning agent library developed by Meta's Applied Reinforcement Learning team. It provides modular PyTorch-based components for building RL agents that handle long-term cumulative feedback, limited observability, sparse rewards, and stochastic environments.

## Use cases
- train reinforcement learning agents in python
- build production RL agents for dynamic environments
- experiment with deep RL algorithms like DQN and SAC
- handle sparse rewards and partial observability in RL
- save and load trained agent state dicts
- research sequential decision-making problems

## When to choose
- you need a modular PyTorch RL library with production ambitions
- your environment has limited observability, sparse feedback, or high stochasticity
- you want serialization of agent components via torch.save/torch.load
- you prefer an MIT-licensed library backed by an active research team

## When to avoid
- you need supervised learning or standard ML pipelines without sequential decision-making
- you want a mature, battle-tested framework like Stable-Baselines3 with a large community
- you need non-Python or non-PyTorch support

## Facets
- artifact type: library
- maturity: active
- function: reinforcement-learning, machine-learning, agent-framework
- domain: reinforcement-learning, machine-learning, artificial-intelligence
- platform: python
- tags: pytorch, meta-ai, production-rl, deep-rl, decision-making, ai-agents

## Member repositories
- facebookresearch/Pearl (main) score 74

## Provenance
- Observed fields: from GitHub, fetched 2026-08-28T04:07:38.576188+00:00.
- Health v2: computed from the inputs above; adoption is never an input.
- Inferred fields (summary, facets, guidance): AI-extracted, prompt v1, taxonomy v1, on 2026-08-29T18:46:52.453810+00:00, confidence not recorded.
  - readme: https://github.com/facebookresearch/Pearl (fetched 2026-08-28T04:07:38.576188+00:00, sha 418515d6a88a)
- Data as of 2026-08-30T08:39:29.467469+00:00.