# aiming-lab/MetaClaw

🦞 Just talk to your agent — it learns and EVOLVES 🧬.

Repository: https://github.com/aiming-lab/MetaClaw
Canonical: https://ross.abutalabs.com/products/metaclaw
Homepage: https://arxiv.org/abs/2603.17187
Language: Python
License: MIT
License Family: permissive
Topics: agent, ai-agent, fine-tuning, llm, lora, metaclaw, online-learning, reinforcement-learning, skill-learning, openclaw, meta-learning, continual-learning, tinker
Last push: 2026-06-07T04:57:05+00:00

## Health v2 (maintenance only)
Score: 68/100 (v2, computed 2026-09-03T02:20:16.233290+00:00)
- activity 86, release rhythm 78, longevity 12
- inputs: {"age_days": 177, "days_push": 87, "days_rel": 144, "gap_med": 1, "n_releases_24m": 8}
- flags: young
- formula: round(0.45*activity + 0.35*rhythm + 0.20*longevity); archived -> min(score, 10)

## Adoption (not part of the score)
Stars 3497, forks 454 (observed 2026-08-28T04:08:07.369860+00:00)

## What it is
MetaClaw is a continual meta-learning framework that lets an LLM agent evolve from real conversations, combining skill synthesis from failure trajectories with opportunistic LoRA/RL fine-tuning during idle windows. It runs as a Python CLI with skills-only, RL, and auto modes and requires no local GPU cluster.

## Use cases
- make my ai agent learn from conversations automatically
- continually fine-tune an llm agent without downtime
- build a self-improving agent with a skill library
- online reinforcement learning for an llm agent
- evolve agent behavior from failure trajectories
- meta-learning framework for deployed agents
- add cross-session memory to my ai agent

## When to choose
- you want an agent that improves continuously from live conversations without manual retraining
- you need skill synthesis plus scheduled LoRA/RL updates without owning a GPU cluster
- you want persistent cross-session memory injected into agent prompts

## When to avoid
- you need a fully static, deterministic agent with no model updates
- you cannot use cloud fine-tuning services like Tinker and want only local inference
- you need a battle-tested production framework rather than a young research project

## Facets
- artifact type: framework
- maturity: active
- function: agent-framework, llm-training, machine-learning, rag
- domain: artificial-intelligence, large-language-models, machine-learning, reinforcement-learning
- platform: python, cli, cross-platform
- tags: continual-learning, meta-learning, lora-fine-tuning, skill-library, online-learning, rl-prm, openclaw, memory, ai-agents

## Member repositories
- aiming-lab/MetaClaw (main) score 68

## Provenance
- Observed fields: from GitHub, fetched 2026-08-28T04:08:07.369860+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 2026-08-29T18:35:51.762464+00:00, confidence not recorded.
  - readme: https://github.com/aiming-lab/MetaClaw (fetched 2026-08-28T04:08:07.369860+00:00, sha 12d5a4eaddc6)
  - homepage: https://arxiv.org/abs/2603.17187 (fetched 2026-08-29T09:30:03.685963+00:00, sha c2117a6edc8e)
  - site_page: https://info.arxiv.org/about/donate.html (fetched 2026-08-29T09:30:03.695062+00:00, sha cca9c3a11c56)
  - site_page: https://info.arxiv.org/about/ourmembers.html (fetched 2026-08-29T09:30:03.698555+00:00, sha 47cbc55ff1de)
  - site_page: https://info.arxiv.org/about (fetched 2026-08-29T09:30:03.700320+00:00, sha a1f16f915a9a)
  - site_page: https://info.arxiv.org/labs/index.html (fetched 2026-08-29T09:30:03.696850+00:00, sha b14a8d05a0ec)
- Data as of 2026-08-30T08:39:29.467469+00:00.