Docs

Creed is your personal context profile. One file that captures who you are: values, goals, work, preferences, constraints, people, health, routines. Any AI you talk to knows you instantly instead of starting from zero every conversation.

It is not a journal, scratchpad, or chat log. The value comes from keeping the profile concise, current, and specific enough that every section actually changes how AI replies to you.

It works the same whether you write code all day or never touch a terminal. The structure is identical for everyone; only the examples shift to match how you work.

Creed runs on a simple loop: you write yourself down once, connected agents read that file before they answer you, and they propose small updates as they learn. You approve the good ones and the profile sharpens over time.

A Creed has ten sections: five always-on core sections everyone fills in, and five optional ones that appear only once you use them. The whole thing is plain Markdown, sized to read end to end in under a minute.

Aim for specific over complete. A short profile that changes how AI replies beats a long one that reads like a resume.

Graph Tags are section references. They tell agents which other sections are nearby context, and they power Nexus, the graph view in the file header.

Typing # in the editor opens a section picker. It becomes a styled chip only when it matches a real visible section, case-insensitively. If it does not match a section, it stays plain hashtag text.

Use a short Graph Tags subsection near the end of each section when it helps. Two to four related sections is usually enough. Do not use Graph Tags for tools, apps, brands, themes, clients, or random labels unless those are actual section names.

Nexus is read-only in v1. It shows visible sections as draggable nodes, colours each node with the section accent, draws tethers from valid section references, ignores fake tags, and shows the section name plus quality score on hover.

The Company plan adds one shared Company Creed on top of your personal one. It is the same structured file, owned by the team instead of a person, so every member's agents read the same company context before they act: how the team works, what it is building, the conventions and constraints that apply to everyone.

Members switch between their personal Creed and any Company Creed they belong to from the workspace dropdown. Roles decide what each person and their agents can do. Owners manage billing and members, admins manage members and content, and members read and propose. Section permissions can loosen or tighten who edits what, and every change is attributed and visible in the team activity view.

Billing is per workspace and owner-only. The Company plan is $129 per month, $999 per year, or $1,999 one-time for lifetime, and each includes 10 seats. Extra seats are $12 per month, $99 per year, or $199 one-time. Every Company plan supports BYOK, and the shared usage allowance is $50 per month (a one-time credit on lifetime).

Creed MCP uses OAuth, so there is nothing to copy. You add the Creed server URL to your agent as a custom MCP connector; the agent opens a browser, you click Allow on the Creed consent screen while signed in to creed.md, and it stays connected. The exact URL is on your Connections page (https://creed.md/mcp on the hosted app).

After connecting, the agent reads Creed once to confirm access, then reads it before meaningful work and proposes narrow updates as it learns. You should not need a second setup prompt.

Almost every connection issue is the OAuth step. These cover the common ones.

Connect Creed to the assistants you chat with. Each one reads your profile before it answers and can propose updates you approve.

Long-running assistants can keep your profile current on their own. Point them at Creed and let them review it on a cadence you trust.

Coding agents read Creed before they plan or write, so they match your stack, your conventions, and how you like to work.

Anything that speaks MCP connects from the server URL over OAuth. Clients that can't speak MCP use the HTTP API instead.

Connected agents read Creed before answering you, let it shape how they reply, and propose narrow updates as they learn new things about you. You approve the good ones and the profile sharpens over time.

Propose an update when you learn something durable about the user, something that would change how a future AI should reply to them, not just a one-time mood or task. The test is: would this make every next AI conversation better?

Most bad proposals are not wrong, they are noisy. If something does not change how a future AI should treat the user, it should not be in the profile.

Each section captures a different kind of context about the user. Good agents aim updates at the section that best matches what they learned instead of dumping everything into one bucket.

Examples are often more useful than abstract rules. These are the kinds of updates Creed should accept and the kinds it should keep out.

When you finish helping the user with something real, ask: did I learn something durable about them? Did anything in the profile look stale or wrong? Only then decide whether to propose an update.

Each section sets its own agent permission, so you can keep part of your profile reference-only and let agents maintain the rest. The mechanics differ per section, but the standard stays the same: only durable, profile-worthy context belongs in the file.

Creed can score how good your profile is, section by section, and surface where to sharpen it. Quality analysis runs on a monthly AI allowance included with your plan, with prepaid top-ups when you need more, or on your own OpenRouter key at cost.

It judges how context is written, never what it is about. A section on work and a section on LEGO are held to the same bar. The only question is whether it helps the next AI know you better.

Once connected, an agent has a focused set of tools for reading and improving your Creed. The flat creed_ tools are the recommended path: each one applies your change directly when that section allows direct edits, or files it as a proposal when approval is on. You never pick the mode; the server does, and every call reports what happened.

MCP is the supported way to connect, and it handles authorization for you. For clients that can't speak MCP, the same capabilities are available over a small HTTP API. Each request sends a bearer token in the Authorization header.