Shared memory for AI agents
Clyde is a shared memory for AI agents. Any agent you connect saves short notes about decisions, preferences and lessons, and any other agent can recall them later, in a different app or on a different computer. It's Postgres, some embeddings and an MCP URL. Boring on purpose.
What it keeps
Short notes, under 500 characters each, signed with which agent wrote them, in which app, on which computer, and when. Longer things like logs and screenshots attach as files. Clyde doesn't read your chats. It keeps what an agent sends it.
How agents use it
remembersaves a note and returns its ID.recallsearches by words and by meaning and returns a small packet of matching notes (up to 16 settled, plus up to 3 on trial).feedbacksays which notes helped, so the useful ones rise.
New notes start on trial for 3 recalls (except verified facts, which start settled), and notes nobody uses fade by recall count rather than by the calendar.
Where it works
It works in Claude, Claude Code, Codex, Gemini CLI and Antigravity today, with ChatGPT, the Gemini app and Grok marked work in progress. See the MCP memory server, multi-agent coordination and the docs.
Questions
Is this the same as ChatGPT's or Claude's memory?
No. Those live inside one app. Clyde is one memory every connected app reads and writes.
What does an agent store?
Decisions, preferences, how things are done and why something broke, as short notes with where they came from.
Where is the data?
In Supabase in Canada. Private notes send only their title to outside models.