AReaL v1.0 trains any existing agent by shadowing its base URL
- Source
- x.com
- Date
🚀 AReaL v1.0 is here! Evolve your 🦞#OpenClaw agents (or any agent) with RL—zero changes to agent required. We add a transparent proxy that shadows your agent's base_url, capturing all agent–LLM interactions for RL training. Just swap the URL and go. ✅ One-click agentic RL for any existing agent—no modifications needed ✅ Full #opencode recipe: training code, data, infra, and models all open ✅ Archon Engine: PyTorch-native 5D parallelism via pure PyTorch; uv sync and go (zero manual compilation) ✅ torch.compile by default: instant 10% performance boost out of the box 📊 SOTA on tau2Bench: 73.0% pass@1 (Airline) / 98.3% (Telecom) #opensource #inclusionAI #RL 📄 Paper: https://t.co/wgAws6D9jR 🐙 GitHub: https://t.co/O1Bw2KwjZr Train your OpenClaw agent:
Context
Adding reinforcement learning to an that's already built and running usually means rewriting its code to plug into a training loop. Ant Open Source, under InclusionAI, says AReaL v1.0 avoids that by inserting a transparent proxy in front of the agent's existing base URL, capturing every interaction between the agent and its as training data without touching the agent's code. The team reports state-of-the-art tau2Bench scores of 73.0% pass@1 on the airline task and 98.3% on the telecom task, and released the training code, data and models.
It follows the same project's earlier v0.5.0 release, which introduced the underlying asynchronous training engine, and the still-earlier v0.2 release.
- AI agent — An AI system that doesn't just answer once but works toward a goal in a loop — taking actions, reading the results, and deciding what to do next.
- LLM — A large language model — the neural network behind tools like Claude and ChatGPT, trained on huge amounts of text to predict what comes next.
Checking sign-in…
Loading comments…