AI product Open source
RDM is an open-source Rust package for building Recursive Decision Models, in which a decision model reads a state and typed questions, returns decisions with probabilities, and selects tools or language-model calls. Selected tools can ask the same questions again about a changed or smaller state, creating recursive sub-calls; the package records answers, observations, termination reasons, and request trajectories. It includes the System One protocol and clients, the RDM loop, optional filtering, fitting, dimensional maps, language-model and recursive-language-model features, a sandbox, arbitration, Map and Arbiter primitives, and a WebAssembly module for constructing workflows involving coding agents, information retrieval, and computer use. It can use stand-in rules without network access, local Laya models through llama.cpp, and hosted Jev or OpenRouter models.
3 uses taken from transcripts — each links to the moment in the video.
Standalone Rust package for building fast, portable recursive decision models
Open-source codebase containing the RDM implementation including the Map and Arbiter primitives
A standalone Rust package implementing recursive decision models, for bringing very fast and portable RDMs to all kinds of systems; suggested for people who want to play around in Rust.
1 in the library.