stratless builds your AI a living model of who you are (what you know, how you think, what you're building), so it stops talking over your head, or under it.
stratless.com · npm · How it works · Privacy
Your coding assistant has no idea who it's talking to. So it only has two registers: silence, or a wall of jargon. stratless gives it the missing third thing, a picture of you, read from the conversations you've already had.
npx stratless init
No account. No API key. No cloud. It reads transcripts already on your disk and borrows the claude you already have to read them. Nothing leaves your machine. All you need: Claude Code installed and signed in, Node 18+, and a few sessions of history.
Run stratless profile and it hands your assistant the picture above. Not a rules sheet you wrote. A model of a person, reasoned from your real history, and it sharpens as that history grows.
Install and the full command list live in cli/README.md.
No model of ours. No server. No training. No inference bill. Four steps, all on your machine:
| Step | What happens |
|---|---|
| 1 · Read | Every session is already on your disk in ~/.claude/projects. stratless walks each one into (what the assistant said → how you reacted) pairs. |
| 2 · Judge | It hands each pair to the claude you already have (claude -p, on your own plan) and asks one question: did understanding transfer, and about what? One line back, cached forever, so each exchange is read once. |
| 3 · Synthesize | It reads the whole stack of those judgments and writes the profile above. |
| 4 · Load | It writes ~/.claude/HUMAN.md and points your assistant's config at it, so your next session starts already knowing you. stratless update refreshes it; stratless stop unloads it. |
If the assistant can't answer honestly, it says nothing: a confidently-wrong profile is the one failure that would end this, so silence always beats a guess.
This thing reads your entire conversation history, so the first question any sensible person asks is "is it phoning home?" The whole tool is in cli/, small enough to read in an afternoon and satisfy yourself that it isn't. That is not a slogan. The line count is the trust argument. At fifty thousand lines nobody checks, and "trust me" is the one thing we're not allowed to say.
Everything runs on your machine. Your conversations, the judgments, and your profile never leave it, not for telemetry, not for "aggregate insight," not ever. There is no server, no account, nothing to sign up for. The only network call is to your own assistant, on your own plan, the same place your code was already going.
The profile is a plain text file. It's yours: load it into any other assistant, read it, or delete it.
cli/ the tool. TypeScript, no runtime dependencies. npm: stratless
web/ stratless.com — Nuxt, no modules, prerendered to static HTML
pnpm install
pnpm test # the CLI's tests
pnpm dev:web # stratless.com, locally
The cli/ is published standalone and must stay dependency-free. Never auto-commit. The tree is left green and uncommitted for a human to review.
MIT. The profile is yours, it's a plain text file, and it's free forever.