Crate agent_framework 

agent-framework

A Rust implementation of the Microsoft Agent Framework for building AI agents and multi-agent workflows.

This is the umbrella crate: it re-exports agent_framework_core and, via cargo features, the providers:

Feature	Crate	Default
openai	agent_framework_openai — OpenAI Chat Completions + Responses API	yes
anthropic	[agent_framework_anthropic] — Anthropic (Claude) Messages API	no
ollama	[agent_framework_ollama] — Ollama (local/remote, OpenAI-compatible)	no
gemini	[agent_framework_gemini] — Google Gemini generateContent API	no
mistral	[agent_framework_mistral] — Mistral AI Chat Completions API	no
foundry-local	[agent_framework_foundry_local] — Microsoft Foundry Local (OpenAI-compatible localhost endpoint)	no
bedrock	[agent_framework_bedrock] — AWS Bedrock Converse API (SigV4-signed)	no
github-copilot	[agent_framework_github_copilot] — GitHub Copilot chat API	no
azure	[agent_framework_azure] — Azure OpenAI (api-key / Entra ID)	no
mcp	[agent_framework_mcp] — Model Context Protocol tools (stdio, HTTP, websocket)	no
a2a	[agent_framework_a2a] — Agent2Agent protocol client	no
declarative	[agent_framework_declarative] — YAML/JSON agents & workflows	no
hosting	[agent_framework_hosting] — serve agents over HTTP (DevUI-style, A2A, OpenAI-compatible)	no
redis	[agent_framework_redis] — Redis chat-message store & context provider	no
mem0	[agent_framework_mem0] — Mem0 long-term memory provider	no
foundry	[agent_framework_foundry] — Azure AI Foundry Responses API chat client + Prompt Agents	no
azure-ai-search	[agent_framework_azure_ai_search] — Azure AI Search memory	no
cosmos	[agent_framework_cosmos] — Cosmos DB NoSQL message store	no
copilotstudio	[agent_framework_copilotstudio] — Copilot Studio agents	no
purview	[agent_framework_purview] — Purview compliance middleware	no

full enables everything.

use agent_framework::prelude::*;

let client = OpenAIChatCompletionClient::from_env("gpt-4o-mini")?;
let agent = Agent::builder(client)
    .name("assistant")
    .instructions("You are a helpful assistant.")
    .build();

let response = agent.run_once("What is the capital of France?").await?;
println!("{}", response.text());

Re-exports

pub use agent_framework_openai as openai;

Modules

agent

Agents: the SupportsAgentRun trait and the concrete Agent.

client

The chat client trait and the automatic function-invocation loop.

compaction

Conversation-history compaction.

error

Error types for the agent framework.

history

Conversation-history context providers.

memory

Context / memory providers.

middleware

Middleware pipelines for agents, chat clients, and function invocation.

observability

Lightweight OpenTelemetry GenAI-style span instrumentation, built on the tracing crate, plus optional GenAI metrics behind the otel-metrics feature.

prelude

Commonly used imports for building agents and workflows.

session

Conversation sessions.

settings

Lightweight settings helpers: a secret-masking string newtype and a precedence-based value loader.

skills

Skills: progressive-disclosure capability packages.

streaming

Utilities for decoding streamed bytes.

tools

Tools: executable functions and hosted-tool markers.

types

Core data model: content, messages, responses, and options.

workflow

Graph-based multi-agent workflow orchestration.

Enums

Error

The primary error type for the agent framework.

Type Aliases

Result

The result type used throughout the framework.