Build knowledge agents without embeddings

Vercel's open-source Knowledge Agent Template enables developers to build AI knowledge agents using a filesystem and bash tools (grep, find, cat) instead of the traditional vector database, embedding model, and chunking pipeline stack. The embedding approach falls short when retrieving specific values from structured data, and its failure mode is silent — the agent confidently returns the wrong chunk with no traceable path from question to answer. Replacing the vector pipeline with a filesystem-backed approach reduced the cost of Vercel's internal sales call summarization agent from ~$1.00 to ~$0.25 per call while improving output quality, as the agent leverages skills it already has — reading files, running grep, and navigating directories. The template is open source, supports pluggable sources such as GitHub repos, YouTube transcripts, and custom APIs, and can be deployed as a web chat app, GitHub bot, or Discord bot, built on Vercel Sandbox, AI SDK, and Chat SDK.

Source: tools

Related