FAQ
Short answers to the questions new users and administrators ask first.
Which SDKs work with Janus?
Anything that speaks the OpenAI API: the official openai Python and JavaScript SDKs, LangChain, LiteLLM, continue.dev, and most IDE assistants. Point the SDK’s base URL at PUBLIC_URL/v1 and use a Janus token as the API key. No other change is needed.
Does Janus store my prompts?
Not for API traffic: on /v1 calls only request metadata is recorded — who called which model, when, token counts, cost, latency, and status. There are three exceptions. Janus Chat keeps your conversations, encrypted, so you can return to them. A troubleshooting session that an administrator switches on captures the bodies of matching requests for a limited time. A Security Gateway violation keeps the matched text, encrypted, for review.
Why does my bill differ from the provider’s invoice?
Janus attributes cost from resolved rate cards: admin overrides take priority over upstream metadata, then exact provider references. Discovery refreshes automatic prices; administrators can set per-field overrides with effective dates. If a rate card lags a provider price change, attribution drifts until the card is updated. Text responses where the provider returned no usage block are estimated from byte counts and marked as such in the log; media responses with no usage are recorded at zero and flagged as a metering gap. Calls on your own personal subscription are not charged to the organization.
What happens when I hit a quota?
The request is refused with policy.quota_exceeded and a reset_at timestamp. By default an in-flight request is allowed to finish; hard-kill quotas cut streams immediately. Retrying before reset_at fails the same way.
Can a team lead manage quotas?
Yes, if an administrator switches on “Team leads may edit quotas” for that team. The lead then manages that team’s quotas from the team page — scoped to that team only, and audit-logged like any admin change. Leads can manage who is on their team either way.
Is a token tied to my account?
Yes. Every personal API token belongs to one person; its usage, grants, and quotas are that person’s. (Service tokens, which administrators create for unattended integrations, belong to no person.) Only the SHA-256 digest of a token is stored, so a lost token cannot be recovered — revoke it and create a new one.