Terminal or Vercel runtime logs
AI_APICallError: 403 model_not_allowed — Model requires privacy mode 'all' or is not in the catalog
- What it means
- Your code asked Tokentide for a model that your key's privacy setting doesn't allow, or used a misspelled model ID. Unlike 503 or 429, 403 `model_not_allowed` should be fixed in configuration, not sent to the fallback.
- Likely causes
- Calling a Claude, GPT or Gemini model using a Tokentide key set to
privatemode (open-weight models only) instead ofall. - Passing a Vercel AI Gateway model ID (with a provider prefix like
anthropic/...) to the Tokentide client instead of Tokentide's catalog ID.
- Calling a Claude, GPT or Gemini model using a Tokentide key set to
- Prevention
- Keep model IDs in one config file as a checked pair (Tokentide ID ↔ Gateway ID) and match your key's privacy mode.
Copy for your agent
Tokentide returned `403 model_not_allowed`. Check `GET /v1/models` to verify our Tokentide and Gateway model ID pair in configuration, and tell me if my key needs privacy mode `all` or if we should pick an open-weight `private` model.
Was this page helpful?