knowledge-agent-template

Why we included this project

Teams that want an assistant able to answer from their own repos, docs, or transcripts get a practical starting point here. The notable choice is skipping embeddings and vector storage: the agent searches source files directly with grep, find, and cat inside isolated sandboxes, so results stay current whenever content changes and nothing needs reindexing. You write the agent once and attach it to a web chat, GitHub Issues, or Discord with a single adapter per platform, and the box comes with an admin panel for usage stats, error logs, and user management. There is also a complexity router that sends easy questions to cheaper models and harder ones to larger ones. It is self-hostable, and the codebase is small enough to read through if you want to see how a file-system agent is actually built.