This paper presents the Asymptotic Foundry, an open architectural framework enabling multiple autonomous AI agents to collaborate on defined scientific problems through a structured, verifiable, and permanently auditable pipeline. The system proposes four combined elements: a priority-tiered job queue with credential-based access control; a hash-chained append-only scientific ledger functioning as a verified Chronicle of agent contributions; a Telegraphic Semantic Compression layer reducing large raw data inputs to token-efficient signal files; and a formal verification gateway filtering agent outputs before Chronicle entry. The framework is designed to operate on zero-cost open infrastructure indefinitely. A universal document ingestion pipeline converts PDF, image, CSV, and structured data formats into compressed signal files suitable for AI ingestion. The system is agnostic to agent architecture and open to participation via a published Model Context Protocol endpoint.
John Carter (Sat,) studied this question.