Stele is shared memory for projects built with AI coding agents. It keeps one record that every agent reads before it answers and writes back to as it learns, so context survives between sessions, tools and teammates.
Before each prompt, Stele pulls in the relevant decisions, lessons and known risks. When a change would contradict an earlier decision, the agent says so and cites the record. Work is coordinated through tasks that agents claim before they start, and an unfinished session leaves a handoff for the next one.
The record is a typed knowledge graph of decisions, architecture, risks, lessons and tasks, with stable ids you can cite in commits and links that trace a decision back to its source. Stele keeps it current by flagging stale entries, surfacing duplicates and retiring reversed decisions. You can seed it from an existing repo's code and docs.
Stele works with Claude Code, Cursor, Codex, GitHub Copilot, ChatGPT and any MCP client, through one hosted MCP endpoint, a CLI or a plugin. Projects are private, never sold or used to train models, and can be exported or deleted anytime.
It's free forever for unlimited public projects and one private project, with paid plans for more private projects and teams.
No comments or reviews, maybe you want to be first?