Introduction to CoAgents (Early Access)

As an industry, we are quickly learning (and re-learning) that AI agents perform best when they work alongside humans. It’s far, far easier to get an agent to excel at specific things, than to perform the given task fully autonomously (see: self-driving cars).

But bridging between humans and AI agents is no small task. Until now.

CopilotKit’s CoAgents provides infrastructure for connecting LangGraph agents to humans in the loop.

Features of a CoAgent

The 4 core features of a CoAgent are as follows:

1. 💨 Stream intermediate agent state
2. 🫱🏻‍🫲🏽 Shared state between the agent and the application*
3. ❓ Agent Q&A (wait for input)
4. 🚘 Agent Steering

Let’s dig into each of these features!

1. Stream Intermediate Agent State

Stream the agent’s intermediate state to the application as the agent executes, so that the user can see work progressing, rather than stare at a loading spinner.

2. Shared State Between Agent & Application

Bi-directional syncing between application state and agent state — to allow an agent and a human to collaborate on a task.

const { state, setState } = useCoagent();

3. Agent Q&A (wait for input)

Agents can easily and intentionally ask the user questions. With support for:

• Text Feedback: Via the chat UI.
• Structured (JSON) Feedback: Via fully custom UI/UX anywhere in your application (chat UI, modals, popups, forms, etc).

Text Feedback

Structured (JSON) Feedback

4. Agent Steering (Coming Soon)

If needed, allow end-users to replay agent execution from a previous checkpoint, as well as correct agent mistakes.