haskie

Curating sources into collections and exposing them over MCP is the right split. The agent reasons, haskie supplies the passages. I run into the same thing: Claude confidently answers from stale docs when the good article is sitting in my bookmarks. How do you handle a source that changes, like versioned framework docs? Re-import on a schedule, or do you pin a version per collection?

I built haskie so my coding agents can work from the sources I trust. I use coding agents every day. They often make mistakes that a source I trust could have helped them avoid. But I was still the one opening the “right book” to catch them. Giving them web search left me checking their answers against my own sources. With haskie, I import those sources once. Its UI lets me curate them and organise them into collections. It connects to Claude and Codex through the Model Context Protocol (MCP), so my agents can consult those collections as they work. Agents can search several aspects of a question at once. They can explore a topic broadly, read deeply into a specific problem, and use what they learn to decide what to investigate next. That gives them a way to discover, bring findings together, and search again as their understanding grows. haskie returns relevant passages from diverse sources and removes repeated material. It supplies the knowledge and the language model does the reasoning. I also care about keeping my taste. I picked these sources because they match how I think: the great article that settles a question, the paper my work rests on. I want those choices to shape how my agents reason and what they build. Which book or paper do you wish your agent had read before it started working on a problem?