Repo

Danus

Runs verifier-gated multi-agent math proof search and accumulates approved claims in a content-addressed fact graph for later reporting or paper writing.

GitHub ↗Jul 7, 2026Paper ↗

  • llm
  • automated-reasoning
  • multi-agent
  • fact-graph
Multi-agent system pipeline showing workers, verifier, shared storage, and orchestration flow
Multi-agent system pipeline showing workers, verifier, shared storage, and orchestration flow

What it is

Multi-agent math proof-search system that only commits verifier-approved claims into a shared fact graph, then renders verified results into reports or LaTeX papers. Orchestrates a swarm of math reasoning agents that prove one claim at a time, stores only verifier-approved facts in a dependency graph, and turns verified results into reports or papers.

Key results

  • Only verifier-approved claims enter truth via the fact graph
  • Real run logged 3,157 verified facts and 8,616 dependency edges
  • Final theorem used a 664-fact supporting closure; chains reached depth 54

Shows that long-form mathematical proof search can be scaled by accumulating only verifier-approved local claims in shared memory.

One useful ML email a week.

Worth your time, not all of your time.

We track new ML work across papers, code, and data, then keep only the pieces worth a closer look. Selected, not collected.

Papers.Data.Code

The Weekly ML Signal

Selected, not collected.

One useful ML email a week: the papers, repos and datasets worth your attention.

Worth your time, not all of your time.