Cecil
A memory layer for AI assistants that tracks what a user said, when, and whether it still lines up with what they are saying now, buildable as a bot's memory backend, a standalone chat app, a Discord bot, a REST API or an MCP server for Claude Code.
Cecil is a memory layer that builds a structured model of a user's world from ongoing conversation: the people they mention, the projects they are working on, the beliefs they hold, and things they said they would do but have not, flagging when a new statement contradicts an earlier one instead of only retrieving matching facts. One codebase runs as several different things: a developer can import it as the memory backend behind a Discord bot, Slack bot, Telegram bot or custom agent with one client call that handles both the chat turn and the memory update, run it as an MCP tool server exposing recall, store, reflect and other operations to Claude Code or Claude Desktop, or call an 11-endpoint REST API from any language.
It requires Qdrant as its vector store, started through Docker Compose, and an LLM endpoint configured through environment variables; a single npm install and npm run dev starts a local web UI for trying it directly, and npm run mcp starts the MCP server instead.
The project is versioned at 2.1 at the point captured.