The memory layer for your AI coding agent
Claude Code is brilliant, until the session ends and it forgets your architecture, your decisions, and every “we don’t do it that way here.” Jeeves captures all of it automatically, keeps your documentation in syncwith the code, and hands the context back at the start of every session — as plain markdown in your repo, never on our servers.
works withClaude Codemore editors coming
Prefer to read it? Step through the same loop yourself.
◇those dim lines are the whole product — you never typed them, and you never will
01
The problem
Claude Code is brilliant for one session and blank by the next. So you have been compensating: re-stating conventions, re-justifying decisions, re-pasting the same context — every single morning. Not because the agent is weak, but because nothing survives the session boundary.
The notes that would fix this never get written, because writing them is a second job nobody asked for.
You have been the memory. That was never supposed to be your job.
field note — “we use pnpm here, not npm” · typed for the third time this month.
02
How it works
capture · while you work
Decisions, patterns, and preferences land in a living knowledge base while you stay in flow. No form to fill, no button to press. Silence is the feature.
▸ noted 14:32 — decision: soft-delete over hard-delete (audit trail)
keep it true · between sessions
Notes drift; work moves on. Jeeves watches the gap — detects when docs no longer match reality, auto-heals broken references, and flags what has gone stale.
▸ healed — docs/auth.md → src/auth/session.ts (path moved)
recall · at session start
At the start of each session — or the moment it becomes relevant — the right context is handed to your AI. No copy-paste. No re-explaining.
▸ recalled — 12 notes · 3 decisions · 1 preference at session start
And what recall surfaces, the next session sharpens — III hands back to I. The loop closes; the memory compounds.
03
Why Jeeves
i.
Built for Claude Code today, more editors coming. Jeeves sits beneaththe agent you already use and makes it sharper every session — a complement, never a replacement. You change nothing about how you work.
cross-session memory·prompt-scored recall
ii.
Everything lives on your machine, as plain files inside your project. Your code never leaves it — signed-in accounts sync only anonymous usage counts, and you can turn that off.
living docs ⇆ code sync
iii.
There is no new window to talk to. Jeeves is the knowledge substrateyour tools stand on — it answers to them, not to a prompt box.
silent decision capture
iv.
Run jeeves --reportand see exactly how much context it recalled for you — a ledger of explanations you never had to repeat. Receipts, not vibes.
value ledger·staleness detection + auto-heal
04
Get started
Two steps, under a minute: install the plugin, then one free sign-in that switches Jeeves on — no keys to copy, ever.
install · 10 seconds
Paste this into Claude Code to add Jeeves. The free sign-in in step 2 switches on capture, recall, and living docs.
sign in · one command
Run one command and Jeeves opens your browser to approve — no keys to copy, nothing to paste. Sign in (the account is free), click approve, and you’re done.
done · every repo
That’s it. One global key now covers every repo you work in — Jeeves wrote it for you, so there’s nothing left to configure.
Tracking a repo separately is optional — run /jeeves:activate <label> inside it to mint a project key inline. No file editing.
Pricing
Jeeves runs inside your repo — every note it takes is a plain file you own, always. It’s free during early access. Later, heavy use will draw on a monthly allowance — but the free tier stays generous, and your files never stop being yours.
Free · early access
$0during early access
Everything Jeeves does — capture, keep it true, recall — with credits on the house while we’re early.
jeeves --report value ledger