A filesystem-based alternative to vector database RAG pipelines for building knowledge agents. Instead of embeddings and chunking, the agent uses bash tools (grep, find, cat) inside isolated Vercel Sandboxes to search files deterministically. This approach reduced cost from ~$1.00 to ~$0.25 per call in a real use case, and makes debugging straightforward since you can trace exactly which commands ran. The open-source Knowledge Agent Template is built on Vercel Sandbox, AI SDK, and Chat SDK, supports multiple platforms (Slack, Discord, GitHub) via adapters, includes a complexity router for cost optimization, and ships with a built-in admin interface.
Table of contents
What the template doesFile-based search with Vercel SandboxChat SDK: one agent, every platformDeep integration with AI SDKBuilt-in admin toolsGet startedSort: