Danus: Orchestrating Mathematical Reasoning Agents with Fact-Graph Memory

2026-07-07 · Source: Artificial Intelligence · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Robotics & Autonomous Systems · Depth: Expert, quick

Summary

Danus is an open-source orchestration system designed to scale and coordinate LLM-based mathematical reasoning agents for research-level problems. Introduced on 2026-07-07, it addresses challenges in parallel proof search and organizing intermediate claims by employing a shared fact graph as its global memory-management mechanism. The system comprises a main agent for planning and coordination, multiple worker agents executing parallel proof searches, and a stateless verifier that validates mathematical claims before their admission into the fact graph. Verified facts are stored with their proofs and logical dependencies, enabling incremental construction of long arguments. The main agent also summarizes proof states, redirects workers, and facilitates human interaction. Evaluated through six research-level case studies in algebraic geometry, singularity theory, and combinatorics, Danus demonstrates its ability to construct detailed mathematical proofs, suggesting fact-graph orchestration is effective for complex, long-horizon research.

Key takeaway

For Research Scientists developing advanced mathematical reasoning systems, Danus offers a robust orchestration model to scale LLM agents for complex, long-horizon problems. You should consider adopting a fact-graph memory architecture to manage parallel proof searches and ensure the reliability of incrementally built arguments. This approach allows your team to tackle research-level challenges in fields like algebraic geometry by systematically organizing verified claims and their dependencies.

Key insights

A shared fact graph effectively orchestrates LLM-based mathematical reasoning agents by organizing claims and proofs.

Principles

• Fact-graph memory organizes evolving proof states.
• Parallel proof search requires coordinated redirection.
• Claim verification ensures reliability for incremental arguments.

Method

Danus orchestrates a main planning agent, parallel worker agents for proof search, and a stateless verifier, all centered on a shared fact graph for verified claims and dependencies.

In practice

• Apply fact-graph orchestration to research-level math problems.
• Scale LLM agents for long-horizon proof construction.
• Integrate human mathematicians via progress reports.

Topics

• Mathematical Reasoning
• LLM Agents
• Fact Graph Memory
• Multi-agent Systems
• Proof Search
• Algebraic Geometry

Code references

• frenzymath/Danus

Best for: AI Scientist, Research Scientist, AI Architect

Related on AIssential

• Artificial Intelligence for Mathematical Reasoning: An Integrated Survey of Language Models, Neuro-symbolic Systems, and Verified Discovery — AI for mathematical reasoning is a rapidly evolving frontier, integrating diverse techniques for verified discovery.
• Advancing Mathematics Research with AI-Driven Formal Proof Search — LLM-driven formal proof search, verified by Lean, can solve open mathematical problems and advance research.
• AtomMem: Building Simple and Effective Memory System for LLM Agents via Atomic Facts — AtomMem enhances LLM agent memory by structuring interactions into atomic facts, hierarchical events, and temporal profiles for efficient, associative retrieval.
• GraphDC: A Divide-and-Conquer Multi-Agent System for Scalable Graph Algorithm Reasoning — GraphDC uses a divide-and-conquer multi-agent system to scale LLM-based graph reasoning by decomposing complex graphs.
• Project NIKA: Unlocking Epistemic Agency in 4-Bit Quantized Models — Constraining LLMs with a deterministic governor can induce "epistemic agency" and superior reasoning, even in small, quantized models.
• ReM-MoA: Reasoning Memory Sustains Mixture-of-Agents Scaling — Reasoning Memory and Curated Diversified Memory Routing enable Mixture-of-Agents architectures to sustain performance scaling with increased depth.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Artificial Intelligence.