Bumblehive

小核心，大轰鸣 | Small Core, Big Buzz

A lightweight Python SDK for building a complete Agent Loop in just a few lines of code.

[](https://pypi.org/project/bumblehive/) [](https://www.python.org/) [](https://github.com/wxhcore/bumblehive/actions/workflows/sdk-ci.yml) [](#local-development) [](https://modelcontextprotocol.io/) [](https://opensource.org/licenses/Apache-2.0)

English | [简体中文](./README_zh.md)

---

Highlights

• Tool-driven agent loop with built-in and MCP-backed tools.
• Streaming lifecycle events for model deltas, tool calls, errors, and final results.
• Session-aware runtime with isolated history and concurrency control.
• Modular Python APIs for configuration, providers, tools, skills, and observability.

Quick Start

import asyncio
import os

import bumblehive

async def main() -> None:
    config = bumblehive.RuntimeArguments(
        model=os.environ["BUMBLEHIVE_MODEL"],
        api_key=os.environ["BUMBLEHIVE_API_KEY"],
        base_url=os.getenv("BUMBLEHIVE_BASE_URL"),
        workspace="./demo",
    )
    runtime = bumblehive.from_config(config)
    result = await runtime.run("Summarize Agent Loop in agent-loop.md.")

asyncio.run(main())

Examples

See [examples](examples/README.md) for independently runnable Runtime, Loop, Provider, Tools, Skills, and Observability examples.

Local Development

The Python SDK, Server, and WebUI development workflow supports macOS, Windows, and Ubuntu Linux. It requires Python 3.11+, Node.js 22.12+, and pnpm 10.33.0. A dedicated Conda environment is recommended:

conda create -n bumblehive_env python=3.11 -y
conda activate bumblehive_env
pnpm run setup

pnpm run setup is the only project setup entry point. It installs every Node workspace dependency from the root lockfile, then uses the active Python interpreter to install the SDK, Server, test, and desktop packaging dependencies before running the core environment check. pnpm run dev repeats that lightweight preflight before starting either process. If a requirement is missing, the command stops with a targeted error. The same command works on Ubuntu Linux and does not require the desktop system toolchain. Set BUMBLEHIVE_PYTHON to an interpreter path to override the active interpreter.

For daily development:

conda activate bumblehive_env
pnpm run dev

This starts the Server on 127.0.0.1:18421 and the WebUI on 127.0.0.1:1420. Use pnpm run dev:server or pnpm run dev:web to run one component, and pnpm test for the Python test suite.

Desktop

The optional desktop workflow currently targets macOS and Windows. macOS additionally requires Rust and Xcode Command Line Tools. Windows additionally requires the Rust MSVC toolchain, WebView2, and Microsoft C++ Build Tools with the Desktop development with C++ workload.

After the shared pnpm run setup, start development or create the installer for the current platform:

pnpm run dev:desktop
pnpm run build:desktop

Both commands verify Rust, Tauri, and the platform toolchain, then automatically package the Python Server as a sidecar before starting Tauri. build:desktop creates an app and DMG on macOS or an NSIS installer on Windows.

Source & license

This open-source MCP server is cataloged on AgentStack and links to its original source — we do not rehost the code.

• Author: wxhcore
• Source: wxhcore/bumblehive
• License: Apache-2.0

Install and usage instructions live in the source repository linked above.