Features

The following are the target features for this package:

• Multi-Model Support
• Create agents from model strings (e.g. openai:gpt-4o) or typed providers (e.g. GeminiProvider())
• Automatically check environment for API key if none is provided (not web compatible)
• Streaming string output via Agent.runStream
• Multi-turn chat/message history support via the messages parameter and the Message class (with roles and content types)
• Typed output via Agent.runFor
• Define tools and their inputs/outputs
• Automatically generate LLM-specific tool/output schemas
• Bring your own provider
• Execute tools with validated inputs
• Multi-media input support via the attachments parameter for files, images, and web content
• Embedding generation with Agent.createEmbedding and cosine similarity utilities
• MCP (Model Context Protocol) server support for integrating external tools from local and remote servers
• Logging support using the standard Dart logging package
• Capabilities capture and reporting
• Model discovery via Provider.listModels() to enumerate available models and their capabilities
• Agentic workflows including multi-step and single-step tool calling
• Firebase AI provider (no API keys!)
• Multimedia output
• Audio transcription
• Tools + Typed Output (that's a little sticky right now)
• More OpenAI-compat providers (local and remote, e.g. Ollama, Gemma, xAI, Groq, etc.)