dartantic_ai 0.5.0 changelog

0.5.0 #

Streaming responses via Agent.runStream and related methods.
Multi-turn chat support
Provider switching: seamlessly alternate between multiple providers in a single conversation, with full context and tool call/result compatibility.

added dotprompt_dart package support via Agent.runPrompt(DotPrompt prompt)
expanded model naming to include "providerName", "providerName:model" or "providerName/model", e.g. "openai" or "googleai/gemini-2.0-flash"
move types specified by Map<String, dynamic> to a JsonSchema object; added toMap() extension method to JsonSchema and toSchema to Map<String, dynamic> to make going back and forth more convenient.
move the provider argument to Agent.provider as the most flexible case, but also the less common one. Agent() will contine to take a model string.

Define tools and their inputs/outputs easily
Automatically generate LLM-specific tool/output schemas
Allow for a model descriptor string that just contains a family name so that the provider can choose the default model.

Multi-Model Support (just Gemini and OpenAI models so far)
Create agents from model strings (e.g. openai:gpt-4o) or typed providers (e.g. GoogleProvider())
Automatically check environment for API key if none is provided (not web compatible)
String output via Agent.run
Typed output via Agent.runFor