Skip to content
Select theme

Resource limits and cost

Resource limits

ResourceLimit
Timeout10 minutes
Memory2 GB
CPU2 shared cores
Concurrent tasks per user5

Limits protect platform stability and keep execution behavior predictable.

Cost model

Porter is free to use. Users pay underlying provider costs:

ServiceCost
Fly Machines~$0.01-0.05 per task
Anthropic/OpenAIPer-token pricing
AmpPer Amp pricing

Design intent

  • Keep workers short-lived and predictable.
  • Let users choose their own model providers.
  • Avoid hidden platform margin on execution.

Cost planning guidance

  • Start with smaller tasks to baseline average runtime.
  • Track provider token usage by task type (bug fix, refactor, test generation).
  • Set team-level guidelines for when to use each agent/model tier.

Concurrency behavior

  • User-level concurrency prevents one actor from saturating worker capacity.
  • Queued tasks should execute in arrival order unless manual prioritization is introduced.

Common overrun causes

  • Issues with broad scope and no acceptance boundary.
  • Repositories with heavy dependency install overhead.
  • Long-running test suites executed as part of every task.

Suggested controls

  • Add issue templates that force clear task boundaries.
  • Keep CI setup efficient to reduce worker idle time.
  • Tune timeout and resource profiles as usage patterns become clear.