← Back to workOpen source · Rust · self-improving agent runtime

Tau

Problem

Most agent runtimes are static: they don't learn from their own execution, lock you into one LLM provider, and offer little operational visibility.

Architecture

A pure-Rust agent runtime with a closed-loop reinforcement-learning cycle, an OpenAI-compatible API gateway, multi-provider LLM support, automatic prompt self-optimization (APO), and an operator dashboard for visibility and control.

What I built

Outcomes

100%
Rust, zero-GC runtime

Stack

RustReinforcement learningLLM gateways