Skip to main content
Chat agents are multi-turn conversational agents. Each chat pairs a scenario’s tools with a model and is served over the A2A protocol. You can chat with them directly on the platform or connect them to other systems via the A2A endpoint. A scenario qualifies as a chat when it declares chat=True or its first argument is an array type (conversation history).

​
How to Use

​
From the Agents Page

  1. Go to the Agents page
  2. Click New Agent and select Chat Agent
  3. Select a chat scenario and model
  4. Your chat agent appears under Your Chat Agents
  5. Click it to open the detail dialog — the right side has an embedded chat where you can try it out directly, with conversation tabs at the top

​
From the Environments Page

  1. Open any environment and load a chat scenario
  2. The “Make Chat Agent” button lets you jump to the agents page

​
Building a Chat Scenario

Use chat=True on the scenario decorator: 
from hud import Environment

env = Environment(name="my-env")

@env.scenario("assistant", chat=True)
async def assistant(
    messages: list[dict] | None = None,
    system_prompt: str = "You are a helpful assistant.",
) -> Any:
    """Multi-turn conversation with tools."""
    prompt_messages = [system_prompt]
    if messages:
        prompt_messages.extend(messages)

    response = yield prompt_messages
    yield 1.0 if response else 0.0
The messages array contains the conversation history. The platform manages this automatically — each turn appends to the history and passes the full context.

​
A2A Integration

Every chat agent exposes an A2A endpoint: 
GET  /agents/{agent_id}/.well-known/agent-card.json
POST /agents/{agent_id}/chat
You can connect chat agents to other A2A-compatible systems, use them as sub-agents, or serve them as standalone endpoints. The A2A connection info is available from the detail dialog.

​
SDK Usage

from hud import Environment
from hud.services import Chat

env = Environment("my-env")
chat = env.chat("assistant", model="claude-sonnet-4-6")

r1 = await chat.send("Hello!")
r2 = await chat.send("Tell me more about that.")

# Serve as A2A endpoint
chat.serve(port=9999)

​
See Also