A shared mental model of your whole system

Your AI sees one file at a time. Peek sees the whole system.

Build the mental model once — then ask anything, here or from Claude, Cursor and Copilot.

Map a repo, free →

◆ no card to start·◆ private by default·◆ $5 free credit

§ 01 — The problem

Most agents read your repo one file at a time, with no model of how it all connects.

without Peek

› read_file('src/index.ts')

› read_file('src/app.ts')

› grep('config')

› read_file('src/lib/util.ts')

› read_file('src/lib/helpers.ts')

› read_file('tests/setup.ts')

› list_files('src/routes')

⋮ 24 more file reads

context burned47,200 tokens

still no model of the system¯\_(ツ)_/¯

with }Peeksame question, agent calls

› peek.map('myrepo')

↳ hierarchical atlas · 1.2k tokens

› peek.ask('where does auth happen?')

↳ reasons over the system map · cites its sources

› read_file('gateway/src/auth.ts')

done.

context used3,400 tokens · 92% less

answer qualityactually correct

§ 02 — The product

Chat with your codebase. Right here.

No login, no install. Pick a question on a famous repo and watch the research team work — or ask your own. Same atlas, same agents you get after signing up.

🔒 codepeekr.dev/facebook/react

The library rendering most of the web — traced across the scheduler, reconciler and DOM.

You've used your free live questions for today.

The recorded runs above are real answers from real runs — keep exploring them. Or index your own repo and ask anything: $5 free, no card.

§ 03 — How it works

Index once. Ask anything, anywhere, forever.

Think of it as the mental model a senior engineer carries in their head — built once, shared by your whole team, and handed to your agents on demand.

Connect a repo

Plug in via the GitHub App. We clone a working copy and watch for pushes — your code never leaves our perimeter.

gh app install · 12s

We build the atlas

Files get summarized into folders, folders into a root map. Each level understands the one below — a model that spans your whole system, version-pinned to every push.

indexing · ~$4 / 100k LoC

Ask it — chat or call

Open codepeekr.dev and chat with your repos directly. Or drop our MCP server into Claude, Cursor or Copilot and let them call it as a tool. Same shared model, same research team, either path.

peek.ask(repo) → research team cheffing up

Drops into

ClaudeCursorCopilotAntigravityCodyOpenAI

§ 04 — What you get

A small toolbox that does one annoying thing very well.

⌂

Hierarchical summaries

File → folder → root. Every level understands the one below, so the model holds your whole system — not just the file in front of it.

◈

Multi-agent reasoning

A director plans the questions. Researchers gather evidence in parallel. A reviewer challenges the draft. You get a reasoned answer, grounded in the code.

↻

Reindex the diff, not the repo

Each push re-summarizes only the files that actually changed — and that's the only spend a reindex costs.

⚿

Private by default

GitHub App scopes, per-repo access, single-tenant clones. Your source never trains anyone's model — yours included.

◇

MCP integration

One remote server, every agent that speaks MCP. Mint a token in Settings, point your client at the URL, and get list_repos and ask_repo as native tools.

claude mcp add --transport http peek \
  https://codepeekr.dev/api/mcp \
  --header "Authorization: Bearer <token>"

⌬

Multi-repo chat

Tick a few repos in the sidebar; ask one question. The team researches across all of them and stitches the answer together.

acme/gateway

acme/billing

acme/web

+ 4 more

Design docs & RFCs

Ask for a migration plan, an RFC, or an onboarding doc grounded in the actual code. Sources cited inline.

draft · rfc-0042Moving auth out of the gateway. Today's flow lives in gateway/src/auth.ts:14…

“

A coding agent is a brilliant reasoner with no mental model of your codebase. A senior engineer carries that model in their head. Peek builds it once — and hands it to whatever you're working in.

— why we built Peek

§ 05 — Pricing

Pay only for what your agent actually uses.

No subscription. Start free, then top up whenever you need more. We pass model cost through at cost + a flat $0.50 per million tokens.

Free

Freeto try

◆ Free $5 worth of tokens

✓Public & private repos
✓Unlimited repos
✓Email support
✓Enough to index a small repo and ask dozens of questions

Start mapping →

pay as you go

Pay-as-you-go

cost + $0.50/ million tokens

◆ Top up any amount, any time

✓Public & private repos
✓Unlimited repos
✓Email support
✓Tokens never expire — always roll over
✓No subscription. No seats. No tiers.

Add tokens →

§ 06 — FAQ

Common questions

Have one not here? Reply to any of our emails — humans answer.

What’s a "token" here, exactly?Same tokens your LLM counts. Most queries: a few hundred to a few thousand.

Indexing a repo spends a one-time chunk of tokens — we summarize every file. After that, asking questions spends tokens proportional to depth.

Do you train on my code?No. Private workspace. No fine-tuning, no sharing.

How does Peek plug into Claude / Cursor / Copilot?MCP server today. Tools: list_repos, ask_repo.

How is this different from Cursor or Claude Code reading my repo?They re-read your repo every session, per developer. Peek maps it once and shares that map.

What about DeepWiki, Qodo or Augment Code?Good tools. Peek is the no-seat, pay-as-you-go way to ask your private repos — here or from your agent.

Can it really answer across multiple repos at once?Yes. Tick the repos, ask once. Per-repo citations in the answer.

If my team adds the same repo, do we each pay to index it?No. Everyone shares one index. Only the first user in the queue pays per reindex.

Give your agent the map.

Free $5 worth of tokens on the house. No card to start.

Start mapping a repo →Read the docs