Docs · everything
in one place.

Pepeview is an AI code-review bot that lives as a GitHub App. When a pull request is opened — or someone says @pepeview on a PR — the frog reads the diff, posts a real, actionable review (bugs, security flags, perf notes), and signs off with a blunt verdict: 🐸 LGTM or 🐸 NGMI.

Usage is gated by the $PEPE token on Base. The token launches separately on Clanker — Pepeview only ever reads the balance, never deploys or mints it.

Install the App from the install page and pick the repos you want the frog to review. The first PR after install triggers the first review automatically.

• Works on individual repos or whole orgs.
• Public repos are reviewed on the Free tier without any wallet setup.
• Private repos require linking a wallet holding $PEPE (see Tiers).

Pepeview ACKs the webhook fast, debounces rapid pushes (rebases, force-pushes), and reviews only the latest head SHA per PR. The pipeline:

PR opened / @mention  →  webhook (HMAC verified)
                          │
            in-process queue · debounced · latest SHA wins
                          │
            ┌─ idempotency: one review per (repo, PR, head SHA)
            ├─ gating: free vs holder · rate limit · private-repo rule
            ├─ diff assembly: skip generated · per-tier byte cap
            ├─ Claude review: persona + rubric (cached system prompt)
            └─ one summary review + inline comments (event: COMMENT)

Comment @pepeview (or @pepeview review) on a PR conversation to re-run the review on the current head. Mentions skip the debounce — the frog reviews now.

Every review returns four things:

• Summary — 2–5 plain-English sentences on what changed and the main risks.
• Inline comments — the most important findings, tied to a specific line and prefixed with a severity tag: [bug], [security], [error-handling], [readability], [perf].
• Verdict — lgtm if safe to merge as-is, ngmi if real issues should be fixed first.
• Verdict reason — one blunt in-character sentence justifying the call.

The rubric, in priority order: correctness/bugs → security → error handling → readability → performance.

Pepeviewis a chain-pilled blue Pepe frog with a half-lidded "I've reviewed worse" attitude — direct, dry, a little mean, but genuinely helpful. The personality lives in the verdict; the actual review stays legible, useful engineering feedback. The persona never roasts the person, only the code.

The entire $PEPE utility for MVP is gating. Holders get the better tier — no staking, no lockups, no funnels. Pepeview reads the balance on Base on every review.

You become a holder when your linked Base wallet holds at least HOLDER_MIN_PEPE $PEPE (whole tokens, scaled by decimals). Balances are cached for 60 seconds.

Two short steps from the dashboard:

• Sign in with GitHub — proves the GitHub identity (only the read:user scope is requested).
• Connect & sign with your wallet — you sign a one-line SIWE message in your wallet (Coinbase, MetaMask, etc). The signature binds githubUserId ↔ wallet without ever touching your private key.

• Free tier: 10 reviews per repo per UTC day. Hitting the limit posts a friendly in-character message and stops.
• Diff size: capped per tier. When a PR exceeds the cap, the most-changed files are reviewed first and the omitted ones are reported in the summary.
• Generated / vendored files: auto-skipped (lock files, dist/, node_modules/, vendor/, minified JS, maps, etc.).
• Synchronize debounce: rapid pushes for the same PR collapse into one review of the latest head SHA.

Pepeview requests the minimum permissions a review needs:

• Pull requests — read & write (to post the review).
• Issues — read & write (PR comments are issue comments).
• Contents — read (to fetch the diff).
• Metadata — read.

The diff is sent to Anthropic to produce the review. Only the changed hunks are sent; full repo contents are never read. No wallet keys, no funds movement, no on-chain writes.

• Uninstall from GitHub → Settings → Applications → Installed GitHub Apps any time. Pepeview drops everything tied to that installation.
• Per-repo opt-out via a .pepeview.yml with disabled: true at the repo root (small follow-up).

The full source ships ready to deploy. The provided docker-compose.prod.ymlruns the api + web + Postgres on a single box. Put Caddy in front for TLS and point your GitHub App's webhook URL at https://api.your.domain/webhooks/github.

docker compose -f docker-compose.prod.yml up -d --build

The api applies pending Prisma migrations on boot. See the project README for the full Hetzner walkthrough, env reference, and how to wire the real Clanker $PEPE address.

Ready to ship a PR for the frog?

Install once. Open a PR. Get a verdict.

Install on GitHub →