Abbenay Python Client

A Python gRPC client for the Abbenay daemon.

Installation

Install from PyPI:

pip install abbenay-client

Note: The pip package is abbenay-client but the Python import is abbenay_grpc:

from abbenay_grpc import AbbenayClient

Cross-platform: works on Linux, macOS, and Windows with Python 3.9+.

Quick Start

import asyncio
from abbenay_grpc import AbbenayClient

async def main():
    async with AbbenayClient() as client:
        async for chunk in client.chat("openai/gpt-4o", "Hello!"):
            if chunk.type == "text":
                print(chunk.text, end="")
        print()

asyncio.run(main())

Event Loop Lifecycle

grpc.aio channels are bound to the event loop that was active at connect() time. If you call asyncio.run() more than once (which creates and destroys loops), call reconnect() in the new loop:

async def preflight():
    async with AbbenayClient() as client:
        await client.health_check()

asyncio.run(preflight())  # loop created and closed

async def main():
    client = AbbenayClient()
    await client.connect()  # auto-detects dead channel, reconnects
    # ... or explicitly: await client.reconnect()

Connecting to a Container

When the daemon runs in a container with TLS (default image CMD), connect via TCP and trust the daemon CA:

async with AbbenayClient(
    host="localhost",
    port=50051,
    tls=True,
    ca_cert="/path/to/ca.crt",
) as client:
    async for chunk in client.chat("openrouter/anthropic/claude-sonnet-4", "Hello!"):
        if chunk.text:
            print(chunk.text, end="")

The container must be started with --grpc-port 50051 (the default Containerfile CMD does this with --grpc-tls) and the port published (-p 50051:50051). Copy or mount <runtime-dir>/tls/ca.crt for client trust.

See docs/CONTAINER.md for full container deployment instructions and the --insecure escape hatch.

Features

• Chat: Streaming chat with any configured model
• Sessions: Create, list, fork, and manage chat sessions
• Models: List available models from all providers
• Tools: Dynamic MCP server registration and tool filtering
• Secrets: Manage API keys via keychain
• Configuration: Get/update daemon config

Requirements

• Python 3.9+
• Abbenay daemon running (abbenay daemon)

License

MIT