Installation

At the end you will have a working way to run the pipeline: the public Docker image (no install at all), docker compose against a checkout, or a local Python environment. You only need one of the three.

Option A — Docker image (nothing to install)

A public image carries the tools, templates, and a default configuration. It is rebuilt automatically on every push that touches the agent.

docker pull ghcr.io/senthilsweb/job-scout
docker run --rm ghcr.io/senthilsweb/job-scout trends

Named jobs the image understands:

Job	What it does	Cost
load	fetch every configured board into ats_posting_raw	free
export	write the trends + full parquet snapshots to exports/	free
report	render the dashboard from the newest export	free
trends	load + export + report in one go	free
match	the resume match sweep — paid, refuses to run unless RUN_PAID_MATCH=yes is set	paid

Anything else is executed as a command (bash, python tools/...), so the image doubles as a toolbox.

To run against your own state on a server, mount a folder and point JOB_SCOUT_CONFIG at a config file that uses absolute paths:

docker run --rm -v /srv/js:/state -e JOB_SCOUT_CONFIG=/state/config.yaml \
    ghcr.io/senthilsweb/job-scout trends

Option B — docker compose (image + your checkout)

From agents/job-scout/ in a repo checkout. The image supplies Python and dependencies; the checkout supplies code, config.yaml, the DuckDB file, and exports/.

docker compose --profile trends up          # boards -> parquet -> dashboard
docker compose --profile match up           # PAID match (reads ./.env)
docker compose --profile shell run --rm shell   # interactive bash

Every service sits behind a profile, so a bare docker compose up starts nothing by accident. The full environment-variable contract is listed at the top of docker-compose.yml and explained in Configuration.

Option C — local Python (for the notebook and development)

cd agents/job-scout
pip install marimo duckdb pyyaml pandas python-dotenv anthropic certifi
marimo edit notebook.py

certifi matters on macOS: without it, Python often has no TLS root certificates and every board fetch fails silently. The tools load it automatically when installed.

The command-line tools need only a subset:

pip install duckdb pyyaml certifi jinja2
python tools/raw_load.py --stats

Secrets (.env)

Copy the template and fill in what you use. Nothing here is needed for the free trends pipeline.

cp .env.example .env

Variable	Needed for
ANTHROPIC_API_KEY	the notebook's optional agentic search mode
RAINFOCUS_PROFILE_ID, RAINFOCUS_COOKIE	loading conference sponsor catalogs
JOBMATCH_API_BASE, JOBMATCH_AGENT_BASE	pointing the paid matcher at a different deployment

.env is git-ignored. Never put secrets in config.yaml or any committed file.

Notes

• The marimo notebook is not in the Docker image — it is an interactive, local surface. Use Option C for it.
• The image is built for amd64 and arm64 (Apple Silicon runs it natively).
• If docker pull says the package is not found, the GHCR package may still be private — it must be flipped to public once in GitHub package settings.

Next: Configuration.