agents 1.0.0

Dart port of the Microsoft Agents AI framework. Build, compose, and orchestrate AI agents backed by any ChatClient — with sessions, streaming, tool use, compaction, evaluation, and middleware pipelines.

agents

A Dart port of the Microsoft Agents AI framework. Build, compose, and orchestrate AI agents backed by any ChatClient — with sessions, streaming, tool use, compaction, evaluation, and middleware pipelines.

C# namespace	Dart layer
Microsoft.Agents.AI.Abstractions	abstractions/
Microsoft.Agents.AI	ai/
Microsoft.Agents.AI.Hosting	hosting/
Microsoft.Agents.AI.Workflows	workflows/

Features

• Provider-agnostic — works with any ChatClient (OpenAI, Azure OpenAI, Anthropic, local models, or a custom stub)
• Stateful sessions — AgentSession persists conversation history, state bags, and conversation IDs across turns; serialisable to/from JSON
• Streaming — first-class Stream<AgentResponseUpdate> support
• Tool use — automatic function-invocation middleware; inject custom approval or retry logic around every tool call
• Composable middleware — AIAgentBuilder chains decorators (logging, telemetry, evaluation, compaction) with a fluent API
• Chat history providers — pluggable ChatHistoryProvider for in-memory or custom persistence backends
• AI context providers — AIContextProvider hooks enrich instructions, messages, and tools before each agent run
• Evaluation — AgentEvaluator with named EvalCheck assertions
• Skills — file-backed and in-memory skill sources with YAML front-matter
• Multi-agent workflows — Executor, Workflow, and MessageRouter for orchestrating agent pipelines
• Hosting — DI / IHost integration for lifecycle-managed agents and workflows

Getting started

Add the package to your pubspec.yaml:

dependencies:
  agents: ^1.0.0

Then fetch dependencies:

dart pub get

You will also need a ChatClient implementation from your AI provider of choice. The extensions package (a transitive dependency) defines the ChatClient abstract type; wrap your provider's SDK in a ChatClient adapter to use it with this package.

Usage

Basic agent — single turn

// Wrap any ChatClient with a system prompt, name, and tools.
final agent = myChatClient.asAIAgent(
  name: 'Assistant',
  instructions: 'You are a concise, helpful assistant.',
);

final session = await agent.createSession();
final response = await agent.run(
  session,
  null,
  CancellationToken.none,
  message: 'What is the capital of France?',
);

print(response.text); // Paris

Multi-turn conversation with a session

AgentSession carries chat history and state across runs so you do not have to manage history manually.

final session = await agent.createSession();

Future<String> chat(String message) async {
  final response = await agent.run(
    session, null, CancellationToken.none, message: message,
  );
  return response.text ?? '';
}

print(await chat('My name is Ada.'));
print(await chat('What is my name?')); // recalls "Ada" from history

Streaming

final session = await agent.createSession();

await for (final update in agent.runStreaming(
  session, null, CancellationToken.none,
  message: 'Write me a haiku about Dart.',
)) {
  stdout.write(update.text ?? '');
}
stdout.writeln();

Composing middleware with AIAgentBuilder

final agent = AIAgentBuilder(innerAgent: baseAgent)
  .use(agentFactory: (inner) => LoggingAgent(inner, logger))
  .build();

Every call to agent.run(...) now passes through LoggingAgent before reaching the underlying ChatClientAgent.

Tool use

final getWeather = AIFunction.create(
  name: 'get_weather',
  description: 'Returns current weather for a city.',
  (Map<String, Object?> args) async {
    final city = args['city'] as String;
    return 'Sunny, 22 °C in $city';
  },
  parameters: [AIFunctionParameter(name: 'city', type: AIFunctionParameterType.string)],
);

final agent = myChatClient.asAIAgent(
  instructions: 'You are a weather assistant.',
  tools: [getWeather],
);

The built-in FunctionInvokingChatClient middleware resolves tool calls automatically before returning the final response.

Additional information

• Source: github.com/jamiewest/agents
• Upstream C# framework: github.com/microsoft/agent-framework
• Docs: learn.microsoft.com/agent-framework
• Issues: please file on the GitHub issue tracker