❓ Question: Integrating ADK with CopilotKit via AG-UI protocol

[rusherman]

Since there are currently few frontend demos available for Google ADK, I’m using CopilotKit in my frontend project.

I was wondering — is there any way to make ADK use the AG-UI message protocol so that it can communicate with CopilotKit?

If possible, could you also provide a small example or point me in the right direction? That would be greatly appreciated.

Thanks in advance!

[SynapticSolutionsAI]

Seconded!

[jsingh-github]

100%. copilotkit is great frontend to connect ai agents. appreciate if google ai team can enable this integration. https://www.copilotkit.ai/

[hangfei]

Right now, we mostly rely on adk-web for UI. Is that sufficient for your need?

[SynapticSolutionsAI]

Not really - CoPilot-Kit / AG-UI offers full application integration.

[MikkelGrey]

https://x.com/CopilotKit/status/1935363243687477434

[hangfei]

are they supporting ADK now?

[SynapticSolutionsAI]

Am confirming with them.

[hangfei]

Here is the FR at CopilotKit someone else posted: CopilotKit/CopilotKit#1856.

[SleeperSmith]

Update. ADK is now supported by AG-UI

https://github.com/ag-ui-protocol/ag-ui/tree/main/typescript-sdk/integrations/adk-middleware
https://docs.copilotkit.ai/adk

No idea if this one's any good. Never used "npx copilotkit@latest create" myself.
https://developers.googleblog.com/en/delight-users-by-combining-adk-agents-with-fancy-frontends-using-ag-ui/

[SleeperSmith]

However

1. That's only python.
2. There are a few important feature missing. Mainly, persistence state restore.

It'd be good if google ADK team can help implement all the integration for all the features.

[rcleveng]

I just started using AG-UI since it seems like a reasonable protocol and better than creating my own (which I was doing before)

Using the AG-UI wrapper on the server for the existing ADK agent works, however I don't think this would work if deployed to Vertex AI Agent Engine

Here's the code:

# Add the ADK endpoint
add_adk_fastapi_endpoint(app, adk_agent_sample, path="/")