Crate agent_framework_foundry 

agent-framework-foundry

An Azure AI Foundry ChatClient and Prompt Agent surface for agent-framework-rs, built on the Responses API (POST {endpoint}/openai/v1/responses) rather than the older Agents threads/runs data plane (removed upstream, along with the azure-ai-agents SDK it wrapped — see this crate’s own history / UPSTREAM_DRIFT.md).

FoundryChatClient does not speak HTTP/SSE itself: it wraps an agent_framework_azure::responses::AzureOpenAIResponsesClient configured for a Foundry project endpoint (Microsoft Entra ID bearer auth or a static API key), with without_api_version applied, since the Foundry v1 GA route is path-versioned ({endpoint}/openai/v1/responses, no ?api-version= query parameter). All request/response conversion, streaming, and error classification are reused verbatim from that client (which itself reuses agent_framework_openai::responses), so wire fidelity comes for free rather than being re-implemented here.

use std::sync::Arc;
use agent_framework_azure::AzureCliCredential;
use agent_framework_core::prelude::*;
use agent_framework_foundry::{FoundryChatClient, FOUNDRY_SCOPE};

let credential = Arc::new(AzureCliCredential::new(FOUNDRY_SCOPE));
let client = FoundryChatClient::with_token_credential(
    "https://my-project.services.ai.azure.com",
    "gpt-4o",
    credential,
);
let agent = Agent::builder(client).instructions("You are concise.").build();
let reply = agent.run_once("Say hi").await?;
println!("{}", reply.text());

Prompt Agents

FoundryAgent realizes a Foundry Prompt Agent client-side: it pairs a FoundryChatClient with a PromptAgentDefinition (name/model/instructions/tools) and runs it over the Responses API through an inner agent_framework_core::agent::Agent. FoundryAgent::to_prompt_agent hands back the definition it was built from, mirroring upstream’s Agent.to_prompt_agent() -> PromptAgentDefinition.

This does not bind to a server-hosted agent by id/name (the Foundry Agents control plane, e.g. AIProjectClient.agents.get(...)) — see the FoundryAgent docs for why that’s a documented extension point rather than something wired up here.

use agent_framework_core::prelude::*;
use agent_framework_foundry::FoundryAgent;

let agent = FoundryAgent::builder(client)
    .name("rust-example-agent")
    .instructions("You are a helpful, concise assistant.")
    .build();
let reply = agent.run(vec![Message::user("Say hi")], None).await?;
println!("{}", reply.text());

// Round-trips the definition the agent was built from.
let definition = agent.to_prompt_agent();
assert_eq!(definition.name, "rust-example-agent");

Structs

FoundryAgent

A SupportsAgentRun that runs a Foundry Prompt Agent over the Responses API.

FoundryAgentBuilder

Builds a FoundryAgent from a FoundryChatClient plus the pieces of a PromptAgentDefinition (name/instructions/model/tools).

FoundryChatClient

A ChatClient for the Azure AI Foundry project Responses API.

PromptAgentDefinition

The serializable definition of a Foundry Prompt Agent: a name, a model, optional instructions/description, and its tools.

Constants

FOUNDRY_SCOPE

The Entra ID scope (audience) for the Azure AI Foundry data plane.