Some of an app’s work is not a server answering requests. A backfill runs once, a report runs every night, a cleanup endpoint needs calling every hour. BWEZE has three ways to run it:
| Runs | Use it for | |
|---|---|---|
| Job | A command, once, on a service’s live release | Backfills, one-time scripts, data fixes |
| Cron job | A command, on a schedule, as its own service | Nightly reports, cleanups, syncs |
| Scheduled URL | A GET to a URL, on a schedule |
Calling an endpoint your app already has |
Run a one-off job
A job runs one command on a service’s live release, in a new instance with the same image and environment variables. It keeps its logs after it ends.
# Runs on the live release and prints its output; the CLI exits with the command's code.
bweze jobs run <serviceId> -- npm run backfill --since 2026-11-01
# Start it and return its ID without waiting.
bweze jobs run <serviceId> -d --time-limit 2h -- ./scripts/reindex.sh
bweze jobs list <serviceId>
bweze jobs logs <serviceId> <jobId> --follow
bweze jobs cancel <serviceId> <jobId>
Because bweze jobs run exits with the code of the command it ran, a script can tell whether the job worked. You can also start a job from the service’s Jobs panel in the dashboard, or from the API:
curl -X POST https://api.bweze.com/v1/services/SERVICE_ID/jobs \
-H "authorization: Bearer bw_sk_..." -H "content-type: application/json" \
-d '{ "command": "npm run backfill", "timeout_seconds": 3600 }'
- A job is stopped at its time limit: one hour unless you set another, from one minute to twelve hours (
timeout_secondsfrom 60 to 43,200). - Up to three jobs run at once per service. A fourth is refused with
429 too_many_jobs. - The service needs a live release (
409 not_deployedotherwise), and hosting must not be stopped for an empty wallet (402 workspace_suspended). - Running or stopping a job needs an API token with
terminal:exec, since the command runs with the service’s secrets. Reading runs and logs needsdeployment:read. GET /v1/services/{id}/jobs/{jobId}?after=<line id>returns the run and up to 2,000 log lines after the one you last saw, so a script can poll for more.
Schedule a command with a cron job
A cron job is a service whose start command runs on a schedule; nothing runs between runs. Pick Cron job as the service type when you deploy the repo, set the start command to what each run should execute, then set the schedule and time limit:
bweze services set <serviceId> --start-command "node dist/report.js"
bweze services set <serviceId> --schedule "0 2 * * *" --time-limit 30m
In the API, the same settings are cron_schedule and cron_timeout_seconds on PATCH /v1/services/{id}.
- A run is stopped at its time limit: one hour unless you set another, twelve hours at most.
- A run never overlaps the one before it. If one is still going when the next is due, the next is skipped and listed as skipped.
bweze jobs list <serviceId>shows past runs and the next ones.- A cron job listens on no port; setting a port on it is refused with
invalid_port.
Write the schedule
A cron job’s schedule has five fields, in UTC: minute hour day-of-month month day-of-week. One that is not five fields is refused with invalid_cron_schedule. In standard cron, * stands for every value of a field, and day of week counts from 0 for Sunday.
| Schedule | Runs |
|---|---|
0 2 * * * |
02:00 UTC every day (04:00 in Kigali) |
0 * * * * |
At the start of every hour |
30 7 * * 1 |
07:30 UTC every Monday |
0 0 1 * * |
Midnight UTC on the 1st of every month |
Standard cron also has ranges, lists and steps such as */15. The BWEZE docs describe only the five fields, so these examples use numbers and *.
Call a URL on a schedule
If your app already has an endpoint that does the work, a scheduled URL calls it with a plain GET on a cron schedule. Add one on the project’s Cron jobs page, or from the CLI:
bweze projects cron create <projectId> -n nightly-report -s "0 6 * * *" -u https://acme.bweze.app/api/report
bweze projects cron list <projectId>
bweze projects cron disable <jobId>
In the API, POST /v1/projects/{id}/cron takes name, schedule and target_url, and PATCH /v1/cron/{id} turns one on or off with enabled.
Each call times out after 20 seconds and does not follow redirects. It carries no signature, so make the endpoint safe for anyone to call, or check a secret in the URL. For work that takes longer than 20 seconds, use a cron job.
How they are billed
Jobs, cron runs and pre-deploy commands are billed like their service while they run: memory at 20 RWF per GiB-hour for the size configured, and CPU at 30 RWF per CPU-hour for what they use. Between cron runs nothing runs, so nothing is billed. The price list has no meter for scheduled URLs; the work happens in whatever the URL points at.
That makes a cron job cheaper than a worker that stays up all day for one task. Here is a report that takes 10 minutes a night, on a service’s 512 MiB and its full 0.5 CPU:
| What | Amount | Rate | Cost |
|---|---|---|---|
| Memory, 512 MiB for 10 minutes a night, 30 nights | 2.5 GiB-hoursat 20 RWF per GiB-hour | 20 RWF per GiB-hour | 50 RWF |
| CPU, the full 0.5 CPU while it runs | 2.5 CPU-hoursat 30 RWF per CPU-hour | 30 RWF per CPU-hour | 75 RWF |
| Before the free allowance | 125 RWF | ||
| Free allowance: 2.5 GiB-hours of app memory | -50 RWF | ||
| Free allowance: 2.5 CPU-hours of app CPU | -75 RWF | ||
| The month | 0 RWF | ||
Computed from BWEZE's published price list when this page was built. Memory is billed only while running; CPU only as used.
The same report in a background worker that is up all month and mostly idle:
| What | Amount | Rate | Cost |
|---|---|---|---|
| Memory, 512 MiB for 730 hours | 365 GiB-hoursat 20 RWF per GiB-hour | 20 RWF per GiB-hour | 7,300 RWF |
| CPU, 2% of one CPU on average | 14.6 CPU-hoursat 30 RWF per CPU-hour | 30 RWF per CPU-hour | 438 RWF |
| Before the free allowance | 7,738 RWF | ||
| Free allowance: 50 GiB-hours of app memory | -1,000 RWF | ||
| Free allowance: 5 CPU-hours of app CPU | -150 RWF | ||
| The month | 6,588 RWF | ||
Computed from BWEZE's published price list when this page was built. Memory is billed only while running; CPU only as used.
The time limit also caps what a stuck run can cost. If the wallet runs out and hosting is stopped, queued jobs wait and new ones are refused with 402 workspace_suspended until a top-up.
The full reference
Jobs and cron in the docs has every command and limit, and Pricing explains how usage is measured.