Common issues
Use this page to narrow a Starfire AI problem to the correct subsystem before changing settings or escalating it.First determine the scope
Ask these questions in order:- Is it one account or multiple accounts?
- Is it one organization or all contexts?
- Is it one model/provider or every model?
- Is it one feature such as FORGE, Research, or Knowledge?
- Does the web app fail while the Developer Platform still works, or vice versa?
- Did the problem begin after a plan, permission, configuration, or release change?
Chat will not send
Check:- the account is still signed in
- the selected model is available
- the conversation is not carrying a broken/unsupported attachment
- plan or organization policy allows the requested feature
- a provider outage is not affecting the route
A model is missing
Effective model availability can depend on:A file exists but Knowledge cannot find it
The file may not be indexed yet, indexing may have failed, the source may be outside the active project/organization context, or retrieval may not consider it relevant to the current query. See Knowledge & RAG troubleshooting for deeper checks.A project is missing
Confirm the active account and organization context. A personal project and an organization-owned project can have different ownership and access rules.Plan changed but access did not
Do not buy the plan again immediately. Check the BillingAccount → Stripe customer → subscription → Starfire plan → entitlement chain. Use Billing troubleshooting.Feature works for another user but not me
Compare plan, organization policy, role, user override, and staged rollout state. A feature flag can intentionally expose an Alpha capability to only part of the user base.Long chat feels slow
Start a new conversation for unrelated work or move durable context into a Project/Knowledge source. Very long histories can increase browser and model-context overhead.When to escalate
Escalate when:- multiple users or organizations are affected
- a critical provider/model route is failing broadly
- billing cannot reconcile
- authentication or permissions are inconsistent with stored state
- background runs are stuck across many accounts
- administrators see unhealthy dependencies
