Providers
Connect upstream AI providers, store provider keys safely, and create models that use those providers.
Providers
Providers are upstream AI services Odock can call: OpenAI, Anthropic, Google Gemini, Azure OpenAI, vLLM, Mistral, Hugging Face, and custom-compatible providers.
A provider is not a model. The provider defines where and how Odock connects. A model defines what applications request and how that maps to an upstream slug.
Provider Hierarchy
The provider and provider key are setup dependencies. The virtual API key is the application credential. Applications never need upstream provider keys.
Provider Types
The organisation UI supports these provider types:
| Provider type | Typical use |
|---|---|
OPENAI | OpenAI chat, responses, embeddings, image, and compatible OpenAI-style traffic. |
ANTHROPIC | Claude models through Anthropic-compatible endpoints. |
GOOGLE | Gemini models through Gemini-compatible endpoints. |
AZURE_OPENAI | OpenAI models hosted on Azure OpenAI. |
VLLM | Self-hosted vLLM inference endpoints. |
MISTRAL | Mistral La Plateforme: chat, Codestral FIM, embeddings, moderation, and OCR. |
HUGGINGFACE | Hugging Face Inference Providers: OpenAI-compatible chat over hundreds of open models and many upstream providers. See Hugging Face. |
CUSTOM | A custom or compatible upstream where your Odock deployment supports the route shape. |
Provider type matters because native endpoints are provider-family aware. For example, an OpenAI-compatible endpoint expects an OpenAI-compatible model. If a request uses the wrong family, the gateway can reject it with a provider mismatch error.
For endpoint behavior, see Endpoints and Native Models call.
Provider Key Security
Odock keeps upstream provider secrets separate from application credentials.
- The provider key is encrypted before storage.
- The full provider key is not displayed again in the UI after saving.
- The gateway resolves and decrypts the provider key only when it needs to call the upstream provider.
- Plaintext provider keys should not be copied into application code.
For deeper security concepts, see Guardrails. For the runtime encryption flow, see Architecture.
Provider Workflows
- Hugging Face
- Activate a provider
- Edit provider details
- Add a provider key
- Add models from a provider
- Update model pricing from the catalog
For the maintained cross-provider list these workflows use, see Model Catalog.
Troubleshooting
| Symptom | What to check |
|---|---|
| Provider does not appear active | Open Providers, clear filters, and confirm the provider card shows Active. |
| Cannot create models | Confirm the provider is active and has at least one provider key. |
| Gateway returns provider mismatch | Confirm the requested model belongs to the provider family used by the endpoint. |
| Upstream authentication fails | Rotate or replace the provider key, then test again. |
| Model calls time out | Review provider timeout and the upstream provider health. |
Continue with Models to configure model records, pricing, policies, and access.