Mintlify has developed ChromaFs, a virtual filesystem that replaces traditional Retrieval-Augmented Generation (RAG) for their AI documentation assistant. The system intercepts standard UNIX commands like grep, cat, ls, find, and cd, translating them into queries against their existing Chroma database. This allows AI agents to explore documentation as if navigating a codebase rather than relying on chunk-based retrieval that could miss information spread across multiple pages or exact syntax buried in results. The approach yielded measurable improvements: session creation time dropped from approximately 46 seconds to 100 milliseconds, and the company eliminated roughly $70,000 in annual infrastructure costs by reusing their existing database rather than provisioning dedicated micro-VMs.
Built on just-bash, a TypeScript reimplementation of bash from Vercel Labs, ChromaFs handles command parsing while translating filesystem calls into Chroma queries. The system stores the complete file tree as a gzipped JSON document within the Chroma collection, enabling ls, cd, and find operations to resolve in local memory without network calls. Access control runs through isPublic and groups fields that prune files from the tree based on user permissions before the agent sees them. For grep operations, ChromaFs uses the database as a coarse filter to identify candidate files, then performs fine-grained regex matching in-memory, allowing large recursive searches to complete in milliseconds.