> ## Documentation Index
> Fetch the complete documentation index at: https://docs.darkbloom.dev/llms.txt
> Use this file to discover all available pages before exploring further.

# Limits and billing

> Concurrent jobs, monthly caps, approval windows, and how a job's charge is held and settled.

## Limits

| Limit | Value | What happens |
| - | - | - |
| Concurrent jobs per installation | 1 to 100, default 2 | More jobs wait for a free slot |
| Monthly cap per installation | Optional, in US dollars per calendar month (UTC) | Settled charges plus open holds count against it; lanes that would not fit fall back to your runners (`monthly_spend_cap`) |
| Approval window | At most 90 days | Lanes fall back after it expires, until it is renewed |
| Lanes per workflow | At most 32 | A malformed line fails the route job |
| Jobs per lane | 0 to 16, default 1 | Darkbloom reserves one slot per job; extra jobs run on Darkbloom only if capacity is free when queued |
| Watchdog wait | 15 to 1440 minutes, default 30 | The watchdog cancels and re-runs a run whose routed jobs never got a runner |

A concurrency limit counts jobs of one installation. The monthly cap counts
settled charges and open holds this UTC month; the console's card shows spend
this month against the cap.

## Pricing

Rates are per minute of each runner class and billed per second. See
[CI pricing](https://docs.darkbloom.dev/ci/pricing.md) for the runner classes, their rates and worked
examples; the live rates are in the sandbox console under **Billing → Rates**.

The Mac's owner earns 90% of each charge.

A job holds funds when it starts and settles to its billable time at the rate
in effect when it started; the rest of the hold is released. The **Runs** page
shows each job's charge (held, settled or released, with the reason), and
**Billing** shows the rate, hold, billable seconds and cost.

## Failed and lost jobs

A job that runs to its own end is billed for its time, whether it passes or
fails: a failing test is the job's own outcome.

A job lost on Darkbloom's side is not charged. If its Mac drops off the network
or its VM is lost, the whole hold is released. Two cases are billed anyway:

* A job that had already finished by itself, and that GitHub had concluded,
  before the loss is billed for the time it used.
* A job that reports success only after its Mac is lost is billed only if
  GitHub concluded it a success. If GitHub failed it instead, the charge is
  released and the job runs again, so you pay for it once.

When a Darkbloom Mac drops off the network while it runs a job, GitHub fails
the job. If Darkbloom operates your approval's watchdog for that lane, it
re-runs such a job once, after the run completes. A re-run is a new job with
its own hold and charge.

A job that cannot be paid for is not started. If a lane was already routed and
Darkbloom then refuses its jobs (a balance too low to hold, a revoked link or
consent, a cap reached meanwhile, or repeated infrastructure failures), the
jobs stay queued on GitHub: cancel the run and re-run it, which routes again.
With the watchdog workflow that happens by itself after the configured wait.

## Spend controls

* **Concurrent jobs** and the **monthly cap** are set per installation with
  **Edit limits** in the console. Limits apply to every repository in the
  installation; new jobs are not admitted past a limit, and running jobs
  finish.
* **Disconnect** stops new jobs and revokes your billing consent. Jobs already
  running finish and are charged.
* To stop routing without disconnecting, set `DARKBLOOM_ROUTING` to anything
  but `on`, or remove the route job.


This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.