MetaClaw

MetaClaw is an agent framework that meta-learns and evolves from live conversations during real-world deployment.

MetaClaw places your model behind a proxy that intercepts interactions from personal agents, injects relevant skills at each turn, and meta-learns from accumulated experience. It supports multiple personal agents including OpenClaw, CoPaw, IronClaw, PicoClaw, ZeroClaw, NanoClaw, and NemoClaw, as well as OpenAI-compatible LLM APIs and Anthropic-compatible endpoints. The platform offers three operating modes for skills injection and reinforcement learning, with weight updates scheduled during idle windows to avoid interrupting active use. It also includes long-term memory to persist facts, preferences, and project history across sessions. No GPU cluster is required because training is handled via cloud-based backends like Tinker, MinT, or Weaver.

MetaClaw at a glance

Best for: Software engineer
Runs on: CLI, macOS, Linux, self-hosted, open source
Pricing: Free
Released: March 2026
Category: AI agent tool
Website: github.com/aiming-lab/MetaClaw

Key features

One-click deployment with metaclaw setup and start commands
Three operating modes including skills-only, RL, and smart scheduler auto mode
Long-term memory for persisting facts, preferences, and project history
Multi-claw support for OpenClaw, CoPaw, IronClaw, PicoClaw, ZeroClaw, NanoClaw, and NemoClaw
Asynchronous serving, reward modeling, and training
Automatic skill summarization after each session
OpenAI-compatible and Anthropic-compatible endpoints

Use it for

Continuously improve an AI agent through real-world deployment
Persist user preferences and project history across multiple sessions
Execute reinforcement learning updates during idle or sleep windows
Automatically inject relevant skills into agent turns

Alternatives to MetaClaw

SkillClaw — Automatically evolves and unifies agent skills across devices and team members.
NemoClaw — Provides managed security, networking, and lifecycle operations for AI agents.
ZeroClaw — Own your AI infrastructure with no subscriptions or cloud dependencies.
NanoClaw — Runs in isolated Docker containers with credential protection via OneCLI.
MateClaw — It secures data and workflows within a self-hosted, audit-ready environment.
OmniAgent — It evolves skills in real-time and dynamically hardens security during task execution.
GoClaw — It offers Go-native performance with multi-tenant isolation and 5-layer security.
NullClaw — High-performance single-binary execution engine for autonomous AI agents.
SelfClaw — It aggregates multiple AI models with automatic fallback and search.
Beeuni — Agents retain context and improve skills over time through a controlled review process.
PhoneClaw — It executes AI agent tasks and system tool calls entirely on-device.
thClaws — Runs locally in native Rust with zero cloud lock-in and multi-provider support.

All MetaClaw alternatives