Crate saorsa_ai 

saorsa-ai: Unified multi-provider LLM API.

Provides a common interface for streaming completions, tool calling, and authentication across multiple LLM providers.

Architecture Overview

┌─────────────────────────────────────────────────────────────┐
│                 Application / Agent Layer                   │
│  (Sends CompletionRequest, receives StreamEvent stream)     │
└─────────────────────────────────────────────────────────────┘
                             │
                             ▼
┌─────────────────────────────────────────────────────────────┐
│              ProviderRegistry (Factory)                     │
│  ProviderKind → ProviderConfig → Box<dyn Provider>          │
└─────────────────────────────────────────────────────────────┘
                             │
               ┌─────────────┼─────────────┬─────────────┐
               ▼             ▼             ▼             ▼
      ┌──────────────┬──────────────┬──────────────┬──────────────┐
      │  Anthropic   │   OpenAI     │   Gemini     │   Ollama     │
      │   Provider   │   Provider   │   Provider   │   Provider   │
      └──────────────┴──────────────┴──────────────┴──────────────┘
                             │
                             ▼
┌─────────────────────────────────────────────────────────────┐
│         Streaming HTTP (reqwest, Server-Sent Events)        │
│  POST /v1/messages → stream of JSON events → StreamEvent    │
└─────────────────────────────────────────────────────────────┘
                             │
                             ▼
┌─────────────────────────────────────────────────────────────┐
│          Message Protocol (vendor-agnostic types)           │
│  Message, ContentBlock, ToolDefinition, ContentDelta        │
└─────────────────────────────────────────────────────────────┘

Provider Abstraction

All providers implement the Provider trait:

• stream_completion: Returns Pin<Box<dyn Stream<Item = Result<StreamEvent>>>>
• Unified event types: StreamEvent::{ContentDelta, ToolUse, Done, Error}
• Model metadata: Context windows, tool support, vision capabilities

Supported Providers

• Anthropic: Claude models with streaming, tool use, vision
• OpenAI: GPT models with streaming, function calling, vision
• Gemini: Google Gemini with streaming and tool use
• Ollama: Local model hosting with OpenAI-compatible API
• OpenAI-Compatible: Generic adapter for compatible APIs (Groq, etc.)

Key Types

• Provider: Core trait for LLM completion providers
• CompletionRequest: Vendor-agnostic request (messages, tools, params)
• StreamEvent: Streaming events (content deltas, tool calls, completion)
• Message: Conversation message with role and content blocks
• ToolDefinition: JSON Schema-based tool specification

Re-exports

pub use anthropic::AnthropicProvider;

pub use error::Result;

pub use error::SaorsaAiError;

pub use gemini::GeminiProvider;

pub use message::ContentBlock;

pub use message::Message;

pub use message::Role;

pub use message::ToolDefinition;

pub use models::ModelInfo;

pub use models::all_models;

pub use models::get_context_window;

pub use models::lookup_by_provider_prefix;

pub use models::lookup_model;

pub use models::lookup_model_by_prefix;

pub use models::supports_tools;

pub use models::supports_vision;

pub use ollama::OllamaProvider;

pub use openai::OpenAiProvider;

pub use openai_compat::OpenAiCompatBuilder;

pub use openai_compat::OpenAiCompatProvider;

pub use provider::Provider;

pub use provider::ProviderConfig;

pub use provider::ProviderKind;

pub use provider::ProviderRegistry;

pub use provider::StreamingProvider;

pub use provider::determine_provider;

pub use types::CompletionRequest;

pub use types::CompletionResponse;

pub use types::ContentDelta;

pub use types::StopReason;

pub use types::StreamEvent;

pub use types::ThinkingConfig;

pub use types::Usage;

Modules

anthropic

Anthropic Messages API provider.

error

Error types for saorsa-ai.

gemini

Google Gemini generateContent / streamGenerateContent API provider.

message

Message and content types for LLM conversations.

models

Model registry for known LLM models.

ollama

Ollama Chat API provider for local inference.

openai

OpenAI Chat Completions API provider.

openai_compat

Generic OpenAI-compatible provider.

provider

Provider trait for LLM backends.

tokens

Token counting and context window management.

types

Request, response, and streaming types for LLM APIs.