job-pipeline
README.md
<p align="center"><picture><source media="(prefers-color-scheme: dark)" srcset="docs/assets/wordmark-dark.svg"><img src="docs/assets/wordmark-light.svg" alt="job-pipeline" width="300" height="56"></picture></p>
<p align="center">A job search pipeline that Claude Code or Codex runs for you. You approve every application.</p>
<p align="center">
<img src="docs/assets/dashboard-light.png" alt="The job-pipeline dashboard for a fictional engineer in San Francisco: jobs ready for review, applications, response and interview rates" width="800">
</p>
<p align="center"><sub>The demo: a fictional engineer searching in San Francisco. Every company is invented.</sub></p>
It finds jobs that match your rules, writes a CV for each one from facts you've confirmed, fills in the application once you approve it, and reads your inbox to see who replied. It never submits an application you haven't approved.
## Let your AI agent set it up
Paste this into Claude Code or Codex:
```text
Set up job-pipeline for me: https://github.com/JoschuaSuenneke/job-pipeline
1. Install Bun if it's missing (https://bun.sh), clone the repo and run `bun install`.
2. Read src/onboarding/init.ts to see the setup questions. Ask me them in plain language, a few at a time.
3. Run `bun src/cli/index.ts init` with my answers piped in, one per line, then run `bun src/cli/index.ts doctor` and fix what it reports.
4. Ask me for my CV file and run `bun src/cli/index.ts import-cv <path>`.
5. Tell me to open Claude Code or Codex in ~/.job-pipeline and say "onboard me".
Don't connect a mailbox, apply to any job or send anything during setup.
```
## Try it
```bash
git clone https://github.com/JoschuaSuenneke/job-pipeline && cd job-pipeline && bun install
bun src/cli/index.ts demo
```
This opens the dashboard with a fictional pipeline in a few seconds, no setup needed. `demo erika` shows a search in Germany instead.
## Use it for your own search
```bash
bun src/cli/index.ts init # targets, commute, pay, rules, your agent
bun src/cli/index.ts import-cv cv.pdf
cd ~/.job-pipeline && claude # or codex
> onboard me
```
After onboarding, you ask for things in plain language:
```
> scout and triage
> tailor the best five
> go for 1 and 3
> check my inbox
```
You need [Bun](https://bun.sh) and [Claude Code](https://claude.com/claude-code) or [Codex](https://github.com/openai/codex). No API key.
## How it works
Claude Code or Codex does the reading and writing. job-pipeline gives it the tools, keeps the records, and enforces your rules.
| Your agent | job-pipeline |
|---|---|
| reads each posting in full and decides whether it fits | pulls jobs from 8 ATS systems (Greenhouse, Lever, Ashby…), 5 remote boards and the German job agency |
| asks you about your experience and writes it down as facts | stores each fact and uses it only after you confirm it |
| writes a CV and cover note for each job | picks the bullets that fit and checks every number, tool and employer against your facts |
| fills in the application form | allows the submit only after your approval, once per job |
| tells you who replied | syncs IMAP, Gmail or Outlook and updates the status of each job |
Everything runs on your machine. Your data lives in `~/.job-pipeline`, never in this repo.
## Rules enforced in code
An agent can ignore an instruction in a prompt, so these rules are checked by the MCP server instead:
- An application goes out only after you approve that job. The approval stores your own words and works once.
- Every number, tool and employer in a CV or cover note has to match a fact you confirmed.
- One application per company, no companies from your blocklist, and a daily limit.
- The agent ticks a consent box only if you allowed that kind of consent, and never when the statement is false.
- It never sends email. At most it puts a draft in your Drafts folder for you to send.
## Configure it for your search
Roles, seniority, years of experience, commute tiers ("San Francisco daily, Palo Alto for 2 office days"), remote regions, pay floor and currency, work authorization, blocklist, consents and mailbox are all set in one config file. It works from the US, the UK and the EU: remote roles are matched to the regions you can work from, and pay is compared only in your currency.
There are two complete examples: [`examples/sam`](examples/sam) (San Francisco) and [`examples/erika`](examples/erika) (Germany). See also [mail setup](docs/mail-setup.md) and [architecture](docs/architecture.md).
## Design choices
It's built to send fewer applications that fit well, not as many as possible. Jobs come only from official sources: public APIs and company career pages. Your data stays on your machine.
## Similar projects
[career-ops](https://github.com/career-ops-hq/career-ops)
[MIT](LICENSE) · [Contributing](CONTRIBUTING.md) · [Security](SECURITY.md)
This server cannot be deployed
Maintenance
ActivityMaintained
ResponsivenessOutdated data