Epi: Towards a Persistent, Self-Improving Agent for Autonomous Research
- Published
- Source
- arXiv
- Paper number
- 1353
- Field
- Research Agents
- arXiv ID
- 2610.24944
Key points
- Épi combines hierarchical memory, versioned branches with backtracking, and independent review to sustain investigations across model sessions and evolve its methods.
- Épi produced 120 record-setting results, comprising 20 numerical bound records across 12 families and 100 construction records across eight families.
- The interval for the n-queens constant became approximately 84-fold narrower, and the rigorous interval for the quartic Feigenbaum attractor dimension became more than 3.6-million-fold narrower.
- Exact dual certificates establish that the Cohn–Elkies linear-programming bound cannot be sharp in dimensions 10 and 11.
- Épi gives a computer-assisted proof of Tuza's conjecture for split graphs with an eight-vertex clique part and at most three triangle-active neighborhood types.
- The lightweight configuration Épi-turbo achieves a 5.4% higher total benchmark score than Codex on 30 FrontierOR tasks at 64% of its model cost.
Paper links
External research summaries. These are not HDATF publications or measured product results.