FAQ¶

Where are the `AddOpenAIProvider()` / `AddAnthropicProvider()` extensions?¶

They no longer exist. Registration is catalog-driven: one call to AddAiProviders() registers every provider from the embedded catalog, and you configure each one through named options keyed by provider ID:

builder.Services.AddAiProviders();
builder.Services.Configure<OpenAICompatibleProviderOptions>("openai", o => { ... });

Why do I get `ai/provider-disabled`?¶

Enabled defaults to true — a registered provider is enabled unless you explicitly set o.Enabled = false (inline or bound from configuration). If you see ai/provider-disabled, an explicit Enabled = false is the cause.

Which providers work without a paid API key?¶

The self-hosted ones: ollama, lmstudio, vllm, jan, localai, textgenwebui, tgi, fastchat. Point BaseUrl at the local server, e.g. http://localhost:11434/v1/ for Ollama. Note that OpenAI-compatible providers still validate that ApiKey is non-empty — set any placeholder value for keyless local servers.

Why does `GetModelsAsync` throw?¶

Check SupportsModelDiscovery first. Providers whose catalog definition has hasModelDiscoveryApi: false throw AiException with code ai/model-discovery-not-supported. See Model Discovery.

Does streaming work for every provider?¶

All four provider implementations implement IStreamingChatProvider, so StreamAsync is always callable — but whether the backend actually serves SSE depends on the service. Check provider is IStreamingChatProvider before calling StreamAsync; fall back to ChatAsync where streaming is not available.

Can I send images or other multimodal content?¶

Yes. Set ChatMessage.ContentParts to a list of ContentPart values (text and image parts). An image part carries an ImageContent, whose source can be a URL, in-memory bytes, a stream, or a local file:

var message = new ChatMessage
{
    Role = EChatRole.User,
    ContentParts =
    [
        new ContentPart { Type = "text", Text = "Describe this image." },
        new ContentPart
        {
            Type = "image_url",
            Image = ImageContent.FromFile("diagram.png")
            // or FromUrl(url), FromBytes(bytes, "image/png"), FromStream(stream, "image/png")
        }
    ]
};

The OpenAI-compatible protocol sends the image as a URL (encoding raw bytes as a data: URI); the Messages API protocol sends a base64 or url source block; the KeyQuery (Gemini) protocol sends an inlineData (base64) or fileData (URI) part. When ContentParts is null or empty, Content is sent as a plain string, so text-only callers are unaffected.

Can I replace the `IAIProviderFactory` implementation?¶

AddAiProviders() registers DefaultAIProviderFactory, which resolves providers from the container. If you need different behavior (key storage, per-call key overrides, fallback order), implement IAIProviderFactory and register it after AddAiProviders() to supersede the default.

How do I know which protocol a provider uses?¶

Query the catalog:

EProviderProtocol protocol = catalog.Get("gemini")!.Protocol;   // KeyQuery

Or read the protocol field in the corresponding ai-providers/{id}.json file — the mapping table is in Wire Protocols.

FAQ¶

Where are the AddOpenAIProvider() / AddAnthropicProvider() extensions?¶

Why do I get ai/provider-disabled?¶