Resto-Bot

Code for a Mastodon bot that draws attention to stale POIs.

Resto-bot reads its candidates straight from a GeoParquet file produced by cosmo — no PostgreSQL/PostGIS required. It picks a random POI that hasn't been edited in over five years, toots a link to it, and remembers what it posted in a plain text log.

Data prep

Build the candidate dataset from an OSM PBF extract with cosmo and the pois.yaml filter. This extracts every amenity=restaurant POI as a point (building polygons are reduced to their centroid) into one GeoParquet file:

cosmo convert \
 --input ~/osm/data/us-latest.osm.pbf \
 --filters pois.yaml \
 --output ./data/pois.parquet \
 --format parquet

Re-run this whenever you want fresher data (e.g. download a new extract and regenerate on a schedule). There is no replication or spatial join step anymore; the place/state shown in each toot is fetched on demand from OSM Nominatim when the bot runs.

Setup

python3 -m venv venv
venv/bin/pip install -r requirements.txt

Configuration is read from environment variables:

• DATA_PATH — the GeoParquet file produced by cosmo; defaults to ./data/pois.parquet
• POSTED_FILE — path to the log of already-posted features; defaults to ./posted.txt
• NOMINATIM_USER_AGENT — identifying string with a contact address; required by Nominatim's usage policy
• MASTO_BASE_URL, MASTO_ACCESS_TOKEN — your Mastodon instance and token; required unless DRY_RUN=1
• DRY_RUN — set to 1, true, or yes to print one post without sending it or updating POSTED_FILE

Docker

Configuration is read from a .env file in the project root, which compose auto-loads for the ${VAR} substitutions in docker-compose.yml. Copy the template and fill it in once — no need to pass settings on the command line:

cp .env.example .env
$EDITOR .env

Build the image:

docker compose build

Seed the data volume from the current Geofabrik United States extract. This downloads us-latest.osm.pbf and converts it to the GeoParquet file used by the bot:

docker compose --profile seed run --rm seed

The seed step is intentionally separate from the bot run because the US PBF is large. Override the extract URL or force regeneration. Keep each of these on a single line — the trailing-\ multi-line form is easy to mangle on paste, which leaves the variable unset:

OSM_PBF_URL=https://download.geofabrik.de/north-america/us-latest.osm.pbf docker compose --profile seed run --rm seed
FORCE_SEED=1 docker compose --profile seed run --rm seed

FORCE_SEED=1 is required to rebuild pois.parquet after changing pois.yaml, since seeding skips regeneration when the file already exists.

Run the bot. With DRY_RUN=1 in .env this prints one post without sending it or touching posted.txt; blank out DRY_RUN and set the Mastodon values for a real toot. Rebuild first if you've changed bot.py since the last image build:

docker compose build restobot
docker compose run --rm restobot

The restobot-data Docker volume stores the downloaded PBF, generated parquet data, and posted.txt.

Running

DRY_RUN=1 NOMINATIM_USER_AGENT='restobot/1.0 (+https://codeberg.org/mvexel/restobot)' venv/bin/python3 bot.py

For a real toot, include the Mastodon settings and omit DRY_RUN:

NOMINATIM_USER_AGENT='restobot/1.0 (+https://codeberg.org/mvexel/restobot)' MASTO_BASE_URL='https://botsin.space' MASTO_ACCESS_TOKEN='...' venv/bin/python3 bot.py

Each run selects one stale, not-yet-posted POI, reverse-geocodes its location, toots about it, and appends its key (e.g. node/123456) to POSTED_FILE.

Crontab it:

NOMINATIM_USER_AGENT='restobot/1.0 (+https://example.org/contact)'
MASTO_BASE_URL='https://botsin.space'
MASTO_ACCESS_TOKEN='...'
0 10 * * * cd /home/mvexel/restobot/ && venv/bin/python3 bot.py >> /home/mvexel/restobot.log