Skip to main content

​
Overview

Providers in Asta represent different AI services that can process chat requests. Each provider has its own configuration, models, and capabilities.

​
Available Providers

Asta currently supports the following AI providers:
  • ollama - Local Ollama instance
  • groq - Groq cloud API
  • google - Google Gemini API
  • claude - Anthropic Claude API
  • openai - OpenAI API
  • openrouter - OpenRouter API

​
List Providers

curl https://api.asta.app/api/providers \
  -H "Authorization: Bearer <your-jwt-token>"

{
  "providers": [
    "ollama",
    "groq",
    "google",
    "claude",
    "openai",
    "openrouter"
  ]
}

​
GET /api/providers

Retrieve a list of all available AI providers.

​
Response

​
providers
array
Array of provider identifier strings. These can be used in the provider parameter when sending chat requests.

​
Provider Details

Each provider implements a standardized interface with the following characteristics:

​
Provider Response Structure

All providers return responses in a consistent format: 
interface ProviderResponse {
  // The AI's response text
  content: string;
  
  // Error information (if any)
  error?: "auth" | "rate_limit" | "model" | "timeout" | "transient";
  error_message?: string;
  
  // Additional metadata
  meta?: {
    model?: string;
    usage?: {
      prompt_tokens?: number;
      completion_tokens?: number;
      total_tokens?: number;
    };
  };
  
  // Tool calls (for function calling)
  tool_calls?: Array<{
    id: string;
    type: "function";
    function: {
      name: string;
      arguments: string;
    };
  }>;
}

​
Error Types

​
auth
error
API key is invalid or missing
​
rate_limit
error
Rate limit exceeded for the provider
​
model
error
The specified model does not exist or is not accessible
​
timeout
error
Request timed out
​
transient
error
Temporary error (connection reset, 500 error, etc.)

​
Provider Configuration

Providers are configured via environment variables or the Asta configuration file. Each provider may require different credentials:

​
Ollama

OLLAMA_BASE_URL=http://localhost:11434
Runs models locally. No API key required.

​
Groq

GROQ_API_KEY=your-groq-api-key
Fast inference for open-source models.

​
Google (Gemini)

GOOGLE_API_KEY=your-google-api-key
Access to Google’s Gemini models.

​
Claude (Anthropic)

ANTHROPIC_API_KEY=your-anthropic-api-key
Access to Claude models with extended thinking capabilities.

​
OpenAI

OPENAI_API_KEY=your-openai-api-key
Access to GPT models.

​
OpenRouter

OPENROUTER_API_KEY=your-openrouter-api-key
Unified API for multiple AI providers.

​
Provider Selection

When sending a chat request, you can specify which provider to use:

​
Default Provider

If you don’t specify a provider (or use provider: "default"), Asta will use the user’s configured default provider. 
{
  "text": "Hello!",
  "provider": "default"
}

​
Specific Provider

You can explicitly choose a provider: 
{
  "text": "Hello!",
  "provider": "claude"
}

​
Provider Capabilities

​
Streaming Support

All providers support both standard and streaming responses:
  • Use /api/chat for complete responses
  • Use /api/chat for token-by-token streaming

​
Vision Support

Providers with vision capabilities can process images:
  • claude - Supports image inputs
  • google - Supports image inputs
  • openai - Supports image inputs (GPT-4 Vision)
Send images using the image_base64 and image_mime parameters: 
{
  "text": "What's in this image?",
  "provider": "claude",
  "image_base64": "iVBORw0KGgoAAAANSUhEUgAA...",
  "image_mime": "image/png"
}

​
Tool/Function Calling

Providers support tool calling for extended capabilities:
  • Web search
  • Web page fetching
  • Code execution
  • File operations
These are automatically enabled based on your request context.

​
Provider Runtime State

Each provider maintains runtime state including:
  • Connection status - Whether the provider is reachable
  • Model availability - Which models are currently available
  • Rate limit status - Current rate limit state
  • Error tracking - Recent errors and their types
This state is managed internally and used to provide better error messages and automatic fallbacks.

Build docs developers (and LLMs) love

Get started for freeTalk to us