Creed
Creed

Your context filefor all agents

One markdown file every AI reads before it answers.

Get Started
creed.md
Creed
Creed
File
Connections
Settings
Sections
SSteve
Steve / Creed
Saved just now
Identity

Founder and designer building Helm, a CI dashboard for small teams. Previously at Stripe, now based in Lisbon.

I think in systems and ship small. I would rather cut scope than miss a date.

Goals

  • Ship Helm v2 public beta by August.

  • Reach 50 paying teams before raising a seed.

  • Publish one essay a month on developer tooling.

Work

Design in Figma, build the app in Next.js, the backend in Go. Deploy on Fly.

TypeScriptGoFigmaLinearFly.io

Preferences

  • Lead with the answer, then the supporting detail.

  • Keep replies tight unless depth genuinely helps.

  • Skip filler, hedging, and over-praise.

Lead with the answer, then the supporting detail. Keep replies tight unless depth genuinely helps. Skip filler, hedging, and over-praise. Default to TypeScript examples; I work in strict mode.
You corrected three JavaScript snippets to TypeScript this week.
Routines

  • Deep work 7 to 11am, no meetings before noon.

  • Review the week every Friday at 4pm.

  • Ship to production Monday through Thursday only.

Deep work 7 to 11am, no meetings before noon. Review the week every Friday at 4pm. Ship to production Monday through Thursday only. Batch code review into a single block after standup.
You batched every review into one block in our last session.
Constraints

  • Never put a call on my calendar on a Wednesday.

  • Do not suggest tools that need a card just to try them.

  • Ask before posting anything to the team Slack.

Why Use It?

AI adoption is exploding. Context is still trapped in the last tool you used.

0.0B+/m

people use standalone AI tools

Each tool starts cold unless your context travels with you.

0M/m

estimated multi-tool AI users

A simple 35 percent estimate across monthly AI users.

0.0T/m

context tokens left behind

Multi-tool users leaving 5,000 useful context tokens behind.

How It Works

The profile your agents read, update, and keep sharp.

Every agent reads it first

Before it answers, any agent pulls your Creed over MCP, so you never re-explain who you are, what you're building, or how you like to work.

Plan my week so I actually ship Helm v2.
Read creed.md
Thought for 4s
Blocking7to11amfordeepwork,nomeetingsbeforenoon,andWednesdaysstaycall-free.IlinedtheweekuparoundtheHelmv2beta.Wantitdaybyday?
Ask anything…

It updates as it learns

When an agent notices something durable, it proposes a precise edit. It lands in your Creed as a diff. Approve it and the section updates in place.

Routines

  • Deep work 7 to 11am, no meetings before noon.

  • Review the week every Friday at 4pm.

  • Ship to production Monday through Thursday only.

proposed·+9−0

And it sharpens over time

Creed scores every section for signal, what's specific and what's thin, so your profile keeps getting sharper without you auditing it.

Identity
…/ 100
SpecificConcreteThinVague
Scoring this section…
+Reads like a real person
/One line stays generic
−Contains a stray line

Review everything or nothing

Approve every agent edit, or let them write directly.

For strategic advice, always includePrefer doing the uncomfortable constraint or the underlying root cause, even when it is not askedwork over asking for permission, and walk through every assumption one at a time. For product feedback, takeunless the timeaction is destructive or hard to explain what users will not care about, not just what sounds good in a demo, and caveat each pointundo. Lead with the context it depends on. When reviewing a plananswer, list every possible objection before giving an opinion, rank them by likelihood, and never commit to a direction until all of them have been written down, debated, and addressed in full. Before shipping anything, restatethen the goal, summarise the alternatives you considered, and ask for sign-off in writingreasoning. DocumentRaise the rationale for every decision in a shared docsingle biggest risk, link the supporting data, and keep a running log of what changed and why so nothing is ever lostnot every objection. When in doubtShip, gather more input, schedule a review, and wait for consensus before moving forwardthen iterate.
You keep choosing to ask rather than act.

You control what gets remembered.

Agents propose updates in real time, but nothing changes until you approve it.

Require approval for agent edits
Review proposed edits first.
BeliefsDirect
Codex·+18−173w ago
Creed's value depends on keeping the file largetight and exhaustivehigh-signal, capturing every passing preferencepruning stale detail so nothing is ever left outit stays sharp and self-improving as it grows, even detailsand trusting that stopped mattering months agoless is more.

Let trusted agents write directly.

Agents can update your Creed without review, keeping your context current as you work.

AI inside the file

Three focused surfaces, each built for a different kind of help.

Identity
…/ 100
SpecificConcreteThinVague
Scoring this section…
+Reads like a real person
/One line stays generic
−Contains a stray line

Analysis

Score every section for signal, weak spots, and what to sharpen next.

Panel
SearchAskAgent
billing
Open billing settings
↵Take me there

Panel

Search, ask, and let Creed draft reversible edits without leaving the file.

?????????

Tab

Tab is in development and will be coming soon.

Get started in minutes

Three steps to a profile every agent can read.

Question 1 of 3
How would you describe yourself?
Founder and designer in Lisbon
Continue

1Describe yourself

Answer a few quick questions and Creed drafts your starter profile.

All agents
One prompt connects them all.

Paste it into any AI. It reads your Creed before every reply.

2Extract your context

Pull the context you've already built across your tools into one profile.

Credits spend
$3.24
30d
AnalysisTabPanel

3Monitor usage

See where your AI allowance goes across Analysis, Tab, and Panel.

Works with your stack

Connect Creed once, then every AI you talk to knows you instantly.

ChatGPT
ChatGPT
Claude
Claude
Grok
Grok
OpenClaw
OpenClaw
Hermes
Hermes
Cursor
Cursor
OpenCode
OpenCode
Devin
Devin
Codex
Codex
Claude Code
Claude Code
Replit
Replit
Whirl
Whirl
v0
v0
Custom
Custom

Questions

Who you are, what you're working toward, how you like AI to talk to you, the people and routines that shape your week, plus any health, accessibility, or hard noes AI should respect. One concise profile, not a journal.

Because it doesn't stick, doesn't cross tools, and you end up repeating yourself. Creed gives every AI the same profile to read before answering, and lets them propose updates as they learn more about you.

Creed connects to agents like Claude Code, Codex, Cursor, and ChatGPT over MCP, and integrates with GitHub for version control. Support for tools like Notion and Obsidian is coming for editing and storage.

Durable things AI learns about you, a sharper preference, a new routine, a goal that shifted. Not session recap, not mood, not generic praise.

No. You can keep agent edits reviewable, or trust them to write directly when you want a lighter loop. The point is control when you want it, not friction by default.

Both. Creed starts as a personal profile, and the Company plan adds one shared Company Creed that every member's agents read, with member roles, an activity view across the team, and admin controls. It starts at $129 per month for 10 seats.

Stop starting from scratch

Every agent you use, already up to speed.

Get Started
Creed

Personal context for your agents.

Fully operational
Product
PricingCompanyExamplesRoadmap
Resources
DocsLearnContextChangelog
Legal
PrivacyTermsStackContact
© 2026 Creed·byhpbrn