Replace repeated introductions with one short file
Creed keeps the durable things an AI should know about you in one Markdown profile. Its sections cover identity, goals, work, preferences, routines, and optional areas such as constraints, people, health, and current context.
The file is meant to stay short enough to read in under a minute. That matters. A sprawling personal knowledge base creates more context for an agent to sift through; Creed is trying to provide a small set of facts that should actually change the answer.
Connected agents can read the profile before meaningful work and propose narrow updates as they learn something durable. Each section can require approval or allow a trusted agent to edit directly.
A useful first connection is two agents and one preference
Start with a harmless preference such as how you want replies structured. Add Creed to two supported clients, ask each one to read the profile, then give both the same planning task. This checks whether the shared context changes the result without filling the file with private history.
Creed uses an OAuth MCP connection. Codex, Claude Code, Cursor, OpenCode, ChatGPT, and other remote-MCP clients are documented. A first-party CLI exposes the live MCP tools from a terminal.
Our safe check stopped before OAuth. The public product demo exposed section
controls for accepting and rejecting proposed changes, and creed-cli status
in a clean configuration directory correctly reported that it was not
connected. We did not create an account or claim that the full two-agent loop
worked.
Self-host it or pay for the managed service
The open-source build is free and includes the editor, MCP connections, and quality scoring. It needs Node, Supabase, several environment values, and a deployment you maintain.
The hosted Personal plan was $12 per month when checked on 6 August 2026. It adds managed auth, storage, sync, backups, and $5 of included AI usage. Company pricing started at $129 per month for ten seats. Creed also supports using an OpenRouter key for model spending.
Treat the profile and connection tokens as sensitive
A Creed can contain goals, relationships, constraints, health information, and other personal details. The hosted privacy policy says Supabase stores backend data, OpenRouter receives relevant profile sections when an AI feature runs, and connected agents read through tokenised endpoints.
The same policy says connection tokens are currently stored in plain text so the service can verify them. Tokens can be rotated, which breaks the old agent connections. This deserves more attention than the ordinary password advice: connect only agents you trust, grant direct edits sparingly, and keep sensitive sections out until the workflow has earned that trust.
Creed says content is not used to train models and provides export and account deletion controls. Self-hosting gives more infrastructure control, but it also makes backups, access control, and updates your responsibility.