Show HN: Stele – A self-maintaining knowledge graph for AI coding agents
10 hours ago
- Stele is a shared memory record for AI coding agents that stores decisions, tasks, and lessons every agent reads before acting and writes back to as it learns.
- It ensures continuity across tools and sessions: agents like Claude Code, Cursor, and Codex all read and write the same project graph, so context is never lost when switching.
- Setup requires only three commands: install the CLI, sign in, and run /stele:start inside the agent to begin automatic reading and writing of the project record.
- The record surfaces context proactively to prevent repeated mistakes, like flagging a risk before a harmful change is shipped.
- Tasks are project-level and claimed atomically to prevent duplicate work, and an operator can manage the queue end-to-end.
- The graph maintains itself: stale entries are flagged, resolved risks retire, duplicates surface, and reversed decisions stop being treated as truth.
- Users can trace decisions back to the original task, document, risk, or person, and ask the project anything via a built-in assistant.
- Data is private per project, never used for training without consent, and can be exported or deleted anytime.
- Stele is invite-only during beta; it works with hosted infrastructure so there is nothing to run, and it starts from existing code to seed the record.