Find, vet, and advance the right roles.
You'll need:
- Node.js 24 or newer: check with
node -v - An AI coding CLI, either one:
- Claude Code:
npm install -g @anthropic-ai/claude-code(claude.com/claude-code) - Codex:
npm install -g @openai/codex(github.com/openai/codex)
- Claude Code:
Then:
npm install -g careerrat
careerrat start claude # or: careerrat start codexmacOS users can also download the signed, notarized desktop app from the latest release.
That sets up your workspace, opens the local app at http://localhost:7777, and
hands you off to the agent. From there you just talk to it.
- Let it onboard you. It asks a handful of questions and builds your profile from the answers: what roles you want, what you'll accept, what you won't, and the real work you've done. If you'd rather kick the tires first, say "set me up with a quick sample profile."
- Paste a job posting, a description copied from anywhere, a link, or the
sample in
examples/sample-jobs/, and say "evaluate this." You'll get a verdict: keep it or cut it, how well it fits, whether the money works, and what to do next. All from an actual read of the posting. - Say "write a résumé and cover letter for this." It builds them from your own evidence and refuses to invent anything.
- Paste a recruiter email and say "draft a reply." It writes the reply and remembers the thread.
- Open
http://localhost:7777and watch the job appear, move through your funnel, and pick up history. Quick local actions work in the app; longer work opens a visible conversation with the owning skill.
One first-run thing that looks broken but isn't: before you've onboarded,
careerrat doctor will report that your setup is incomplete and list
candidate/*.yml files to create. That's expected. Onboarding fills them in.
careerrat next # the one thing worth doing next
careerrat doctor # check your setup is healthy
careerrat update # pull the latest code; your data is untouchedIf the local app is already running, the update relaunches that recorded CareerRat process on the updated code. Unrelated processes are never stopped; CareerRat selects another loopback port instead.
The local app comes up with careerrat start. To run it on its own:
careerrat tracker # snapshot tracker.json for recovery
careerrat tracker-dev # serve http://localhost:7777 with live reloadUseful flags on start: --no-agent (workspace + local app only),
--no-dashboard, --agent <name> (override with a compatible agent command),
--port <n> (default 7777).
git clone https://github.com/CodesWhat/careerrat
cd careerrat
npm install
npm run hooks:install
npm link
careerrat start claudeCareerRat is a job-search workspace that runs on your own machine. You tell it what you're actually looking for. It reads real job postings and tells you which ones are worth your time, writes applications from things you've genuinely done, drafts your recruiter replies, preps you for interviews, and keeps track of where everything stands.
No account, no CareerRat server, no telemetry. CareerRat never phones home, and your files stay on your machine. What does go out: your AI CLI talks to its own provider to do the work, same as any other task you'd give it, and the app fetches public resources like job postings and company logos from the services that host them. The packaged desktop app also checks GitHub once a day for a newer release and shows an in-app notice, nothing more; it never downloads or installs anything on its own, and it can be turned off in Settings. See privacy for the details.
Most job tools match keywords, then spray. CareerRat won't write a single line of a cover letter until it has read the whole posting and checked it against what you said you want: your comp floor, your location, your dealbreakers. Jobs that don't clear that bar, it tells you to skip.
And it won't lie for you. Every claim in a tailored résumé traces back to something you told it about your own work. If you didn't do it, it doesn't get written.
CareerRat is an agent runtime. The CLI sets up the workspace and serves the local app, but the job-search work happens inside your agent, reading a set of skills that tell it how each step is done. That's why you talk to it in plain language instead of memorizing subcommands. The first run detects supported AI CLIs and explains the available choices.
The rule underneath all of it: no tailoring, no applying, until the job has passed a real read of the posting. Titles and keywords are triage, not truth.
Same skills for anyone. A nurse, an engineer, and a driver each answer onboarding their own way and get the same loop.
- Onboarding: a conversation, not a form. Produces your targets, comp floor, evidence bank, honesty boundaries, and writing style.
- Finding jobs: builds searches from your targets, finds boards and company career pages worth watching, dedupes, drops dead links, and triages what's new.
- Vetting jobs: reads the full posting and judges it against your actual constraints before anything gets written.
- Honest applications: résumés, cover letters, and short answers built only from your evidence bank, with a check that blocks anything half-finished.
- Applying: fills portal forms for you, defaults to letting you hit submit, pauses at CAPTCHAs.
- Recruiter comms: drafts replies, follow-ups, scheduling, and negotiation, and keeps the whole thread.
- Interview prep: packets tailored to who you're talking to, a story bank grounded in your real work, and live coaching for comp conversations.
- Outcome tracking: records what happened, notices when your results say your strategy needs a rethink, and tells you.
- Research: company intel and comp benchmarks, kept firmly separate from your résumé claims so web findings can never launder into fake credentials.
- Dashboard: stat cards, funnel, active pipeline from sourced through offer, per-job detail, follow-up reminders, and table / board / calendar views. Tokyo Night and Gruvbox themes, light or dark.
- Memory: lessons from each application compound, so it gets sharper the longer you use it.
- Roadmap: version themes and what's next
- Sources strategy: how job sources get curated
- Architecture and AGENTS.md: how the skills and the agent contract fit together
Bugs and feature requests: GitHub Issues.
Questions, ideas, and show-and-tell: GitHub Discussions.
Chat: CodesWhat Discord.