Providers and Models

A provider is an AI service (OpenAI, Anthropic, Google, etc.). A model is a specific model offered by that provider (gpt-4o, claude-sonnet-4-20250514, gemini-2.0-flash).

GoAI uses a factory pattern: each provider package exports a Chat() function that returns a model implementing the provider.LanguageModel interface.

import "github.com/zendev-sh/goai/provider/openai"

model := openai.Chat("gpt-4o")

The model is then passed to GoAI functions like StreamText or GenerateText:

result, err := goai.GenerateText(ctx, model,
    goai.WithPrompt("Explain quantum computing in one sentence."),
)

Model Interfaces

GoAI defines four model interfaces in the provider package.

LanguageModel

Text generation and tool calling.

type LanguageModel interface {
    ModelID() string
    DoGenerate(ctx context.Context, params GenerateParams) (*GenerateResult, error)
    DoStream(ctx context.Context, params GenerateParams) (*StreamResult, error)
}

Used with GenerateText, StreamText, GenerateObject, and StreamObject.

Models may also implement the optional CapableModel interface to declare capabilities. Use provider.ModelCapabilitiesOf(model) to query safely.

EmbeddingModel

Vector embeddings from text.

type EmbeddingModel interface {
    ModelID() string
    DoEmbed(ctx context.Context, values []string, params EmbedParams) (*EmbedResult, error)
    MaxValuesPerCall() int
}

Used with Embed and EmbedMany.

ImageModel

Image generation from text prompts.

type ImageModel interface {
    ModelID() string
    DoGenerate(ctx context.Context, params ImageParams) (*ImageResult, error)
}

Used with GenerateImage.

VideoModel

Video generation from text, images, or frame references.

type VideoModel interface {
    ModelID() string
    DoGenerate(ctx context.Context, params VideoParams) (*VideoResult, error)
}

Used with GenerateVideo.

Capabilities

Models that implement the optional CapableModel interface expose their capabilities. Use provider.ModelCapabilitiesOf to query safely (returns zero-value if not implemented):

caps := provider.ModelCapabilitiesOf(model)
if caps.ToolCall {
    // safe to pass tools
}
if caps.Reasoning {
    // model supports extended thinking
}

The ModelCapabilities struct includes:

Field	Type	Description
Temperature	bool	Accepts temperature parameter
Reasoning	bool	Extended thinking / chain-of-thought
Attachment	bool	File attachment support
ToolCall	bool	Tool / function calling
FileUpload	bool	Remote file upload support
InputModalities	ModalitySet	Supported input types (text, image, etc.)
OutputModalities	ModalitySet	Supported output types

Provider Configuration

Providers expose authentication and customization options, but option names vary by provider:

// Explicit API key
model := openai.Chat("gpt-4o", openai.WithAPIKey("sk-..."))

// Dynamic token source (OAuth, service accounts)
model := openai.Chat("gpt-4o", openai.WithTokenSource(myTokenSource))

// Custom HTTP client (proxies, logging, URL rewrite)
model := openai.Chat("gpt-4o", openai.WithHTTPClient(myClient))

Many providers read credentials from environment variables when explicit auth options are omitted (for example OPENAI_API_KEY, ANTHROPIC_API_KEY). Some providers use provider-specific auth options (for example Bedrock uses AWS credential options instead of WithAPIKey/WithTokenSource).

Provider List

Provider	Import Path	Factory Functions
OpenAI	github.com/zendev-sh/goai/provider/openai	Chat, Embedding, Image
Anthropic	github.com/zendev-sh/goai/provider/anthropic	Chat
Google	github.com/zendev-sh/goai/provider/google	Chat, Embedding, Image, Video
Azure	github.com/zendev-sh/goai/provider/azure	Chat, Image
Vertex AI	github.com/zendev-sh/goai/provider/vertex	Chat, Embedding, Image
Bedrock	github.com/zendev-sh/goai/provider/bedrock	Chat, Embedding
Mistral	github.com/zendev-sh/goai/provider/mistral	Chat
xAI	github.com/zendev-sh/goai/provider/xai	Chat
Groq	github.com/zendev-sh/goai/provider/groq	Chat
DeepInfra	github.com/zendev-sh/goai/provider/deepinfra	Chat
OpenRouter	github.com/zendev-sh/goai/provider/openrouter	Chat
DeepSeek	github.com/zendev-sh/goai/provider/deepseek	Chat
Fireworks	github.com/zendev-sh/goai/provider/fireworks	Chat
Together	github.com/zendev-sh/goai/provider/together	Chat
Cohere	github.com/zendev-sh/goai/provider/cohere	Chat, Embedding
Cerebras	github.com/zendev-sh/goai/provider/cerebras	Chat
Perplexity	github.com/zendev-sh/goai/provider/perplexity	Chat
MiniMax	github.com/zendev-sh/goai/provider/minimax	Chat
RunPod	github.com/zendev-sh/goai/provider/runpod	Chat
Cloudflare	github.com/zendev-sh/goai/provider/cloudflare	Chat, Embedding
FPT Cloud	github.com/zendev-sh/goai/provider/fptcloud	Chat, Embedding
NVIDIA NIM	github.com/zendev-sh/goai/provider/nvidia	Chat, Embedding
Ollama	github.com/zendev-sh/goai/provider/ollama	Chat, Embedding
vLLM	github.com/zendev-sh/goai/provider/vllm	Chat, Embedding
Compat	github.com/zendev-sh/goai/provider/compat	Chat, Embedding
Requesty	github.com/zendev-sh/goai/provider/requesty	Chat
Opper	github.com/zendev-sh/goai/provider/opper	Chat
Runcrate	github.com/zendev-sh/goai/provider/runcrate	Chat
Cheaper Inference	github.com/zendev-sh/goai/provider/cheaperinference	Chat
llama.cpp	github.com/zendev-sh/goai/provider/llamacpp	Chat, Embedding

The compat provider works with any OpenAI-compatible API. Pass a custom base URL:

import "github.com/zendev-sh/goai/provider/compat"

model := compat.Chat("my-model",
    compat.WithBaseURL("https://my-api.example.com/v1"),
    compat.WithAPIKey("my-key"),
)