Connect AI model Providers: device, account and authorized access
Choose where a task executes and which model it uses. One selection interface keeps each connection’s ownership and authorization source explicit.
Understand the three sources
- Device: connections configured on the execution computer and managed under its configuration permissions.
- Account: private Providers managed by your Server account. Selecting another computer or team does not copy their keys.
- Authorized platform access: models and budgets explicitly granted by an administrator, retaining their original authorization and usage source.
Match the actual API protocol
Ternilo supports OpenAI Chat Completions, OpenAI Responses, DeepSeek Responses, native Gemini and Claude Messages. Match the endpoint to its actual protocol; a similar model name does not make two interfaces equivalent.
Enter the API root URL rather than a complete chat/completions or responses request path. Add the real model ID or fetch the service’s model directory if available. Context size, output limits and reasoning settings should reflect the service’s actual capabilities.
Configure and test your first connection
- Open Settings → Models locally, or My models on the Server, and choose the device or account configuration target.
- Add a Provider with the endpoint, correct protocol and any required API key.
- Add or fetch models and check their IDs and capacity settings.
- Explicitly select the source and model in the session. Test a simple request without tools before testing project access.
Troubleshoot a failed request
- 401 / 403: check the key, account permissions and model grant.
- 404: check the API root and protocol; avoid duplicate request paths.
- model not found: verify the exact upstream model ID.
- Incompatible response: confirm the protocol actually implemented by the endpoint.
- Timeout: check connectivity from the execution service before changing timeouts.
Shared access is not a shared API key
Seeing a model directory does not reveal its key or authorize forwarding a device Provider to another computer. Team access to models and project resources is granted separately.
Ternilo does not silently fall back to a same-named model if a source goes offline or a grant is revoked. Select an available source explicitly. Actual costs depend on the chosen model service.
Common questions
Can I use a local model endpoint?
You can configure a reachable endpoint with a compatible protocol. Verify tool calls, image support and reasoning settings against the actual service.
How is a Provider different from a model?
A Provider defines the endpoint, protocol, credentials and model directory. A model is a specific model ID under that connection; one Provider can expose several models.
Does choosing another computer switch my model?
Execution placement and model bindings are separate. Selecting a computer does not automatically copy Providers from another device or account.