Resource limits and cost
Resource limits
| Resource | Limit |
|---|---|
| Timeout | 10 minutes |
| Memory | 2 GB |
| CPU | 2 shared cores |
| Concurrent tasks per user | 5 |
Limits protect platform stability and keep execution behavior predictable.
Cost model
Porter is free to use. Users pay underlying provider costs:
| Service | Cost |
|---|---|
| Fly Machines | ~$0.01-0.05 per task |
| Anthropic/OpenAI | Per-token pricing |
| Amp | Per 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.