AI 'Starships' Claim 27 Open Math Proofs, Each Formalized in Lean 4

Star Fleet Math, a Mac desktop app built by Colin Snyder, orchestrates up to 20 agentic harnesses called ‘starships’ in parallel, each running its own GPT-5.6 instance on a dedicated 60-vCPU server and attacking a separate open problem. The whole system is written in TypeScript and Bun. Every starship can burst to thousands of vCPUs or H100 GPUs for sharded search, query what the author claims is the largest indexed corpus of Lean 4 theorems and lemmas via embeddings and a Chroma vector database, pull arXiv papers and GitHub repos through Firecrawl, and lean on a preloaded sandbox of SAT/SMT solvers (CaDiCaL, kissat, Z3), computer algebra systems, and full Rust, CUDA, and Lean toolchains. Proofs are checked by a Claude Fable-based verifier agent, then escalated to the human author over iMessage, and verified lemmas are folded into a persistent dependency graph so results compound across runs.