Getting Started¶

At the end you will have run job-pilot on your machine — first free, then with real scoring — and know what each run costs.

Path 1 — see the data it works on (5 minutes, nothing to install)¶

job-pilot's input is public. Open the trends dashboard or query the parquet straight from any DuckDB shell:

SELECT count(*) FROM 'https://raw.githubusercontent.com/senthilsweb/ai-agents/main/agents/job-scout/data/ats_raw_trends.parquet';

Path 2 — run the tests (5 minutes, no secrets)¶

cd agents/job-pilot
python3 -m venv .venv && .venv/bin/pip install -e ".[dev]"
.venv/bin/pytest -q          # 42 tests, no network, no cost

Path 3 — a real run (needs secrets, small LLM cost)¶

Copy .env.example to .env and fill it — see Configuration for every value.
Set RUN_PAID_MATCH=1 in .env. This is the paid-call switch: the pipeline refuses to call the matcher API without it.
Dry run first. It does everything — real job delta, real scoring, real PDFs — except sending the email:
```
.venv/bin/python run.py --dry-run --baseline trends/20260714
```
Open runs/<date>/digest.html and the PDFs next to it.
Full run. Drop --dry-run and the digest arrives by email.

--baseline names the dated snapshot to compare against (a trends/YYYYMMDD git tag). Without it, the run uses yesterday's tag — in CI, the tag of the last successful run.

Cost note: a typical day has 3–10 new matching jobs; each costs one matcher API analysis. If the delta looks wrong (more than max_jobs_per_run jobs), matching is skipped for that run — zero paid calls, reported in the digest's Failures box — and the rest of the run (including the email) completes normally.

Next: Configuration