Skip to content

Provider support

The provider registry has 210 entries. They are not all equal.

Level What it means Count in 2.2.0
Curated Configuration written and maintained by KODA: endpoint, authentication, region note, defaults, list prices where published. Supported: we fix problems with these. 18
Catalogue Imported from a public provider catalogue so you can try them. Best-effort support. 192
KODA tested A curated provider and model that passed the KODA tested benchmark as a resident: tool use, policy adherence, continuity and Japanese instructions. 1 model; 10 pending
Provider Where requests are processed (provider’s own statement)
Sakura AI Engine (default profile) Japan
Ollama Your machine
Any OpenAI-compatible endpoint Wherever that endpoint is hosted
Azure OpenAI The region you choose per resource
IBM watsonx.ai The region you choose per account (Tokyo, jp-tok, by default)
Mistral AI European Union
OpenAI Global, US by default
Anthropic Global, US by default
Google Gemini API Global; region not selectable on this API
Groq, Together AI, Fireworks AI, xAI Global, US by default
OpenRouter Depends on the upstream model
DeepSeek, Moonshot AI, Z.AI China (international endpoints for some)
MiniMax Singapore

Region statements come from each provider’s documentation and are recorded in the registry with their source. Check your contract with the provider before sending regulated data; for personal data, turn on the context guard and prefer a provider in your own jurisdiction or a local model.

Every release runs the full test suite on Linux, macOS and Windows. Provider adapters are tested with simulated responses; live provider accounts are not part of release testing. Live benchmark runs are added to the KODA tested catalogue as they complete, with a signed report.