1. Home
  2. Kai - AI Assistant
  3. Python Client

Kai Python Client

The Kai Python Client is an official Python library for interacting with the Kai API programmatically. It provides async support, SSE streaming, a command-line interface, and comprehensive type safety through Pydantic models.

Features

  • Command-line interface for quick interactions without writing code
  • Real-time streaming — see responses as they arrive
  • Conversation management — maintain context across multiple messages
  • Tool approval flows — review and approve Kai actions programmatically
  • Full API coverage — chat, history, and feedback

Installation

uv add kai-client

Using pip

pip install kai-client

Prerequisites

The Kai Python Client requires a Master Token to authenticate with the Keboola API. Standard tokens with limited permissions will not work. You can create a Master Token in your project’s Settings → API Tokens.

Quick Start

import asyncio
from kai_client import KaiClient

async def main():
    # Auto-discover the Kai API URL from your Keboola stack
    client = await KaiClient.from_storage_api(
        storage_api_token="your-master-token",
        storage_api_url="https://connection.keboola.com"  # Your stack URL
    )

    async with client:
        # Start a new chat
        chat_id = client.new_chat_id()

        # Send a message and stream the response
        async for event in client.send_message(chat_id, "What tables do I have?"):
            if event.type == "text":
                print(event.text, end="", flush=True)
            elif event.type == "tool-call":
                print(f"\n[Calling tool: {event.tool_name}]")
            elif event.type == "finish":
                print(f"\n[Finished: {event.finish_reason}]")

asyncio.run(main())

Command-Line Interface

The package includes a kai CLI for quick interactions without writing code.

Setup

Set your credentials as environment variables:

export STORAGE_API_TOKEN="your-master-token"
export STORAGE_API_URL="https://connection.keboola.com"

Commands

kai ping                          # Check server health
kai info                          # Show server version and info
kai chat                          # Start an interactive chat session
kai chat -m "List my tables"      # Send a single message
kai chat --auto-approve -m "..."  # Auto-approve tool calls
kai history                       # View recent chat history
kai get-chat <chat-id>            # Get full chat details
kai delete-chat <chat-id>         # Delete a chat

In interactive mode, type your messages and press Enter. Type exit, quit, or press Ctrl+C to end.

Tool approval: When Kai calls a write tool (e.g., update_descriptions, run_job, create_config), the CLI pauses and asks you to approve or deny. Use --auto-approve to skip this prompt.

For more usage examples (non-streaming chat, conversations, tool calls, tool approval, error handling), see the client README.

Use Cases

Integrating Kai into Data Apps

The Kai Python Client can be embedded into Keboola Data Apps to provide AI-powered chat interfaces for your end users.

Important: Data Apps are automatically provisioned with a Storage API token, but this built-in token is not a master token and is insufficient for the Kai Client. You must pass a Master Token explicitly — for example, via a secret environment variable in your Data App configuration.

Python/JS Data Apps

You can integrate Kai into Python/JS Data Apps today using the Kai Python Client directly. A dedicated plugin with ready-made patterns will be available soon to simplify the setup.

Streamlit Data Apps

The kai-streamlit plugin provides patterns and working code for building Streamlit Data Apps with an integrated Kai chat interface. It handles the async bridge between Streamlit’s synchronous model and the KaiClient’s async API, streaming responses into Streamlit containers, tool approval flows with interactive Approve/Deny buttons, and session state management across Streamlit reruns.

To get started, install the dependencies:

pip install kai-client streamlit

Then use the run_async bridge pattern to call KaiClient from Streamlit:

import asyncio
from kai_client import KaiClient

def run_async(coro):
    """Run an async coroutine from sync Streamlit code."""
    loop = asyncio.new_event_loop()
    try:
        return loop.run_until_complete(coro)
    finally:
        loop.close()

See the plugin repository for a complete working example with streaming, tool approval, and suggested action buttons.

Kai via CLI with Claude Code

The kai-cli plugin lets AI coding assistants like Claude Code interact with your Keboola project through the Kai CLI. Install the plugin to give Claude the ability to query data, manage configurations, run jobs, and troubleshoot issues — all through natural language in your terminal.

Installation

Download the plugin to your Claude Code plugins directory:

mkdir -p ~/.claude/plugins

curl -L https://github.com/keboola/kai-client/archive/refs/heads/main.tar.gz | \
  tar -xz --strip-components=2 -C ~/.claude/plugins kai-client-main/plugins/kai-cli

Or clone and link for development:

git clone https://github.com/keboola/kai-client.git
ln -s "$(pwd)/kai-client/plugins/kai-cli" ~/.claude/plugins/kai-cli

Once installed, ask Claude to “use kai” or “help me with kai cli” to activate the skill. Claude can then run kai chat, kai history, kai ping, and other CLI commands on your behalf.

Resources