We built Dexicon because there's invaluable context in AI coding sessions that disappears the moment you close the tab. Architectural decisions, debugging rabbit holes, the "why we did it this way" - gone.

Dexicon captures sessions from Claude Code, Cursor, Codex, and others, then makes it all searchable via MCP. You can also upload sessions manually along with relevant docs. It extracts atomic pieces of context into a knowledge graph - for V1, that means completed tasks and debugging/root-cause analyses, the non-trivial stuff that helps when someone hits the same issue a few weeks later.

It's designed to be useful for solo devs who want searchable insights into their own sessions, but scales to teams as a way to solve the tribal knowledge problem.

We're pre-seed with a handful of paying customers. The developers we've been working with have surprised us with use cases we didn't anticipate: encoding team best practices, speeding up onboarding for new teammates, and generating optimized agent instructions from their own session history.

Now we're opening up access to more users and would love feedback from HN community.