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

# Limits & Quotas Reference

> Understand the categories of Starfire AI limits, where effective values come from, and how to diagnose limit-related failures.

# Limits & quotas reference

Starfire limits are configuration-driven. The effective value can depend on the deployment, plan, organization policy, user override, workload, model, and current Alpha rollout.

Because those values can change independently from the documentation, this page defines the **limit categories and precedence model** instead of hardcoding numbers that can become stale.

## Effective limit precedence

```text theme={null}
Platform safety maximum
        ↓
Plan default
        ↓
Organization policy
        ↓
Resource/workload limit
        ↓
User or credential override
        ↓
Effective limit
```

A narrower limit can reduce access even when a plan would otherwise allow more.

## Limit categories

| Area          | Typical limits                                                                         |
| ------------- | -------------------------------------------------------------------------------------- |
| Chat          | context size, output size, tool calls, attachment count                                |
| Files         | file size, file count, storage, supported formats                                      |
| Projects      | project count, members, storage, attached resources                                    |
| Knowledge     | source count, source size, chunks, indexing concurrency, query result count            |
| Research      | source count, query count, depth, runtime, credits                                     |
| FORGE         | workspace size, file count, runtime, concurrent builds, repair attempts, artifact size |
| Agents        | tool access, model access, run budget, concurrent tasks                                |
| Automation    | schedules, watcher frequency, workflow duration, approvals, concurrent executions      |
| Developer API | requests per window, concurrent runs, daily credits, build concurrency                 |
| Billing       | included credits, hard ceilings, seats, storage, retention                             |

## Find the current value

Use the product surface closest to the workload:

* account or billing UI for personal plan allowances
* organization administration for team policy
* Developer Portal for API quotas
* FORGE UI for build-related limits
* Control Center for platform/operator configuration

## Soft vs hard limits

A **soft limit** can warn or throttle without permanently blocking the account.

A **hard limit** rejects new work after the threshold is reached until the limit resets or an authorized configuration change occurs.

## Limit-related troubleshooting

When a request is rejected because of a limit:

1. Identify the resource/workload.
2. Confirm the active personal or organization context.
3. Check the applicable plan entitlement.
4. Check organization policy.
5. Check explicit user/application overrides.
6. Check whether a hard credit ceiling or concurrency limit is active.
7. Retry only if the limit is temporary or resettable.

<Note>
  A provider can also impose its own upstream limit. Starfire can surface or absorb some provider throttling, but provider limits and Starfire product quotas are different layers.
</Note>