Skip to content

OpenAI chatkit backend support for Google ADK based agentic applications

License

Notifications You must be signed in to change notification settings

ksachdeva/adk-chatkit

Repository files navigation

OpenAI chatkit support for Google ADK

This repository provides the backend support for openai/chatkit-js (https://github.com/openai/chatkit-js) for Google ADK based agentic applications.

It uses and extends openai/chatkit-python (https://github.com/openai/chatkit-python) by providing

  • ADKStore that wraps BaseSessionService
  • A function (stream_agent_response) that translate ADK events into chatkit events
  • Provides support to render widgets
    • See examples/backend/src/backend/agents/facts/_tools.py::get_weather
  • Provides support for making calls to client tools.
    • Client tools typically run in browser
    • See examples/backend/src/backend/agents/facts/_tools.py::switch_theme

TODO:

  • Support for attachments / artifacts

Install

uv add adk-chatkit

Running examples

Make sure you open this repository in vscode devcontainer and all dependencies will be setup for you

# At the root of the repository
# fill in your configuration / settings
cp .env.example .env

There is one backend and one frontend that hosts 4 agents (chatkit servers) and their corresponding user interface.

# Run the backend
uv run poe run-example-backend
# Run the frontend
uv run poe run-example-frontend

Usage

See examples for full usage

class FactsChatKitServer(ADKChatKitServer):
    def __init__(
        self,
        store: ADKStore,
        runner_manager: RunnerManager,
        settings: Settings,
    ) -> None:
        super().__init__(store)
        agent = _make_facts_agent(settings)
        self._runner = runner_manager.add_runner(settings.FACTS_APP_NAME, agent)

    async def _adk_respond(
        self,
        thread: ThreadMetadata,
        item: UserMessageItem | None,
        context: ADKContext,
    ) -> AsyncIterator[ThreadStreamEvent]:
        if item is None:
            return

        if _is_tool_completion_item(item):
            return

        message_text = _user_message_text(item)
        if not message_text:
            return

        content = genai_types.Content(
            role="user",
            parts=[genai_types.Part.from_text(text=message_text)],
        )

        agent_context = ADKAgentContext(
            app_name=context.app_name,
            user_id=context.user_id,
            thread=thread,
        )

        event_stream = self._runner.run_async(
            user_id=context.user_id,
            session_id=thread.id,
            new_message=content,
            run_config=ChatkitRunConfig(streaming_mode=StreamingMode.SSE, context=agent_context),
        )

        async for event in stream_agent_response(agent_context, event_stream):
            yield event

Examples applications

There are 4 example applications (3 are ported from https://github.com/openai/openai-chatkit-advanced-samples)

Facts & Guide

  • Shows Fact Recording
  • Displays Weather using Widget
  • Theme Switching

http://localhost:5173/guide

Weather widget preview

Customer Support

  • Airline Reservation Management
  • Change Seat
  • Add bags

http://localhost:5171/customer-support

Customer Support preview

Knowledge Assistant

  • Answers questions based on files and vector store
  • Shows files and citations

http://localhost:5171/knowledge

Knowledge preview

Widget Gallery

  • Shows various widgets and how to process actions

Knowledge preview

About

OpenAI chatkit backend support for Google ADK based agentic applications

Resources

License

Stars

Watchers

Forks

Packages

No packages published

Contributors 2

  •  
  •