PR Radar Download
Local-first · macOS menu bar app

Code review that runs on your Mac.

Bring your own provider. Your credentials stay in the Keychain; optional cross-Mac archives are end-to-end encrypted. It still greps the repo that calls you, the ticket, the errors and the flags before it judges.

We send the macOS download link to your inbox, which is how we confirm the address is really yours.

14read-only tools Claude, Kimi or local No API key to request Your keys, never ours

No key to request

Your company won't give you an API key. PR Radar doesn't need one.

The AI writes code faster than anyone can review it - and it is still your name on the approval. Reviewing that flood should not begin with a losing fight against Security for a key you will never be handed.

  • It drives Claude Code, not an API. PR Radar runs claude -p - the same CLI you are already cleared to use. Your existing Claude Code or Kimi login is the entire setup.
  • Nothing new to get approved. No API key to request, no SDK to whitelist, no ticket with IT or Security. The access battle most engineers quietly lose never starts.
  • Still yours to override. Prefer a raw API key, or an offline Ollama model that phones nobody? Both still work. Claude Code is simply the path that needs no permission.

Why the others miss it

A diff cannot answer the question

The bugs that reach production are rarely wrong inside the diff. They are wrong about something the diff does not contain.

  • The caller in another repository. The signature changed, this repo compiles, and the service that calls it lives somewhere else.
  • The requirement nobody checked. Clean, tested, reviewed - and not what the ticket asked for.
  • The error it is already throwing. This exact path has been failing in production since Tuesday.
  • The flag that is already on. "Behind a flag" is only reassuring until you check for whom.
  • The fix that changes nothing. The override is gone and the diff reads like a fix - but the default it now falls back to, in a file this pull request never opens, is still the broken one.
PR Radar answers all five by going and looking, with read-only tools, before it writes a word.

Why not CodeRabbit?

Bot reviewers work for the repo. This one works for you.

CodeRabbit, Copilot review and the rest are bots on somebody else's server: they read each diff in isolation, on their machines, with your code. PR Radar is built the other way around.

A bot reviewer
  • Reads the diff, alone
  • Runs on their servers, with your code and their key
  • Comments on every pull request, for everyone
  • Priced per seat, per month
PR Radar
  • Greps your whole organisation, mirrored locally - plus the ticket, errors, metrics and flags
  • Runs on your Mac, on your own key; code and tokens never leave it
  • Reviews what was asked of you, and posts only when you say
  • Cents per review on your own API key, with a hard ceiling - or free with a local model

The reviewer

It investigates before it judges

Every pull request gets a Review button. A real agent reads the change, searches every mirrored repository in your organisation for whoever calls what changed, and follows the work into the ticket, your errors, metrics, flags and docs before it forms an opinion. Because the whole organisation is mirrored, that search is exhaustive: a review that comes back with nothing has looked everywhere, so its silence is a result rather than a blind spot.

prr-diffprr-read prr-checksprr-pr prr-reposprr-git prr-grepprr-recall prr-ticketprr-docs prr-errorsprr-metrics prr-flagsprr-cloud
  • Read-only by construction. Those fourteen tools are its entire reach. Write and edit do not exist, and read-only is enforced by the tools, not requested in the prompt.
  • Your stack, in the Keychain. Jira or Linear, Sentry or Datadog, Grafana, LaunchDarkly, AWS: paste a read token and that tool appears. Credentials live in the macOS Keychain and never reach a transcript.
  • Claude, Kimi or local - and it never approves. Run it through Claude Code, Kimi or a local Ollama model that sends nothing to any provider. Posting is off by default, and approval is a separate switch again, because an approval can be what merges.
  • It remembers across pushes. Review again after a fix and it knows what it flagged last time - so it tells you what you closed and what is still open, instead of handing you the same list rescored.
The PR Radar iPhone transcript for acme/checkout#412: the agent reads the diff, is granted its read-only tools, runs prr-grep 'submitCharge(' across every repository and finds the caller in acme/orders, reads ticket ACME-482, and ends with Request changes.
The PR Radar iPhone review of acme/checkout#412: verdict Changes, a plain-language summary explaining how a request id regenerated inside the retry loop lets a retried charge slip past the idempotency guard and double-charge, then a blocker finding citing src/payments/retry.ts:88, with Post and Review again at the bottom.
The review it hands back: a verdict, a plain-language summary, and findings that each cite a file and a line - with the transcript that got there a tap behind.

Yours, wherever you are

Done when you look. With you when you leave.

Automatic review watches what people ask of you by name - never the team firehose - and starts the moment the request lands. The queue you open is a queue of answers.

  • Every moment, on your lock screen. Review requested, queued, reviewing, the one-word verdict, a colleague asking for another look - pushed as they happen, with the full review and transcript one tap away.
  • Start, stop, post and argue - from your thumb. Dispatch a review, kill a runaway one, post a verdict behind its confirm, or open the chat that already holds the diff, the findings and the ticket. Export any of it as Markdown.
  • Sealed end to end. Phone and Mac pair by camera and speak only to each other's keys. The relay between them carries ciphertext it cannot read, and Apple sees none of it either.
  • Any pull request at all. Somebody pings you in Slack about a PR nobody assigned? Paste the link, from the menu bar or the phone.
92
reached a team you belong to
3
a person asked you by name
The PR Radar iPhone app showing the review queue: an In flight section with acme/checkout#412 reviewing now (ticket ACME-482, checks failing) and acme/api-gateway#233 queued, then a reviewed acme/billing#77 requesting changes and more below.
Real screenshot. The queue you open is a queue of answers - every verdict at a glance, and Stop, Review and Post a thumb away.

What inference costs you

About a dollar, and a ceiling you set

What a review costs your provider on your own key - real numbers from the author's own ledger across dozens of reviews on a 200-repo organisation, not what we charge and not a projection.

$1.04
a typical review, start to verdict - your change read against the whole org
$3.72
the most a single deep, org-wide trace has ever cost
  • A ceiling per review, that you set. The engine stops the run the moment it is hit, not after the bill arrives. Conservative out of the box; raise it for the deep infra changes worth the dig.
  • A rolling daily ceiling too. Set a cap over a rolling 24 hours; reaching it genuinely stops automatic review until it resets.
  • Or nothing at all. Point it at a local model with Ollama: no token leaves your Mac, and nobody bills you.
  • Always visible. The widget carries what the last 24 hours cost, against the ceiling, on your desktop.

Your code, your keys

Everything runs on your Mac. Nothing runs on ours.

Reviews run on your machine with your credentials. Tokens live in the macOS Keychain. Reviews are read-only until you explicitly post; PR Radar writes its private state, transcripts and repository mirrors only to your Macs.

  • Every run keeps receipts. The verdict, every finding, and the full transcript: every command, every refusal, every cent. Rereadable months later.
  • Secrets are never recorded. Tools receive credentials as process environment; transcripts name keys, never values.
  • You can read the prompt. Every instruction PR Radar sends is in Settings, in full, and you can edit it.
  • The widget cannot leak. It ships with no network entitlement and is never handed a token: it draws a snapshot, and its Review button asks the app.
The PR Radar desktop widget on a dark Mac desktop: the pull requests a person asked you to review by name, each with author, repo, PR number, CI status and the agent's verdict badge.
The desktop widget draws a snapshot and holds no token. Its Review button asks the app; it cannot reach the network itself.

Pricing

Free to review. Paid to roam.

Your spend cannot blow up. A hard ceiling per review and a daily ceiling per account, both yours, both enforced inside the run itself. Each commit is reviewed once - a rebase with identical changes carries its verdict for free - and only pull requests that asked you ever dispatch. The bill is boring by design.

The local reviewer is free forever. Pay only when you want it off your desk - on your phone, on every Mac, reviewing on its own.

Free
$0forever

Local manual reviews. No account required.

  • The full local reviewer: 14 read-only tools
  • Cross-repo search across your whole org
  • Claude, Kimi or a local model
  • Review any PR by hand, on your Mac
  • Multiple GitHub accounts, local history and transcripts
Coming soon
Team
Later

For the whole team.

  • Shared config across the team
  • The admin console
  • Org-wide review policy
  • Priority support

~$25-40 / seat

What you get, at a glance

Free is the complete reviewer on each Mac. Pro adds unattended work and one secure account across your devices.

CapabilityFree · $0Pro · $18/mo
Review on this Mac
Manual AI reviewsAny pull request, full verdict and findingsIncludedIncluded
14 read-only toolsCross-repo code, tickets, errors, metrics and flagsIncludedIncluded
ModelsClaude, Kimi, or local OllamaYour providerYour provider
GitHub accountsMultipleMultiple
History and transcriptsStored locallyLocal + encrypted sync
Automation and everywhere access
Automatic requested reviewsRuns unattended when someone asks youNot includedIncluded
All your MacsLive running status, completed history and transcriptsNot includedIncluded
Offline account archiveEnd-to-end encrypted; PR Radar cannot read itNot includedIncluded
iPhone app and push alertsNot includedIncluded
Team LobbyRead-only live review dashboardNot includedIncluded
Terms
PR Radar subscriptionFree forever$18 monthly or $180 yearly
TrialNot needed14 days · no card
Model inferenceNo token markup from PR RadarNot includedNot included

Team features - shared configuration, admin controls, organization policy and priority support - are planned, not part of today's Pro subscription.

The subscription pays for PR Radar's automation and remote features. Model inference is not included: reviews use your own provider account or local model, and PR Radar never marks up a token. Cross-Mac links are end-to-end encrypted, and Pro also keeps an end-to-end encrypted account archive so completed history remains available while the Mac that ran it is offline. PR Radar's service cannot decrypt that archive.

Get PR Radar

Two minutes to your first verdict

Drag to Applications, paste a GitHub token, and your queue is already being read. The next review request lands answered.

We send the macOS download link to your inbox, which is how we confirm the address is really yours.

macOS 15 or laterApple Silicon & Intel Signed and notarized1.9.53-94d571cc