Install

hiveloom 1.1.0 is released and available through Python package indexes. It requires Python 3.11 or newer.

For the hiveloom command, install it as an isolated uv tool:

uv tool install hiveloom

When hiveloom is a dependency of a Python project instead, add it to that project:

uv add hiveloom
# or: uv pip install hiveloom
# or: pip install hiveloom

Verify the install and explore the contract without making a single API call:

hiveloom --version
hiveloom schema --annotated
hiveloom catalog tools

To pin a release exactly, use uv tool install "hiveloom==1.1.0" or uv add "hiveloom==1.1.0".

Development UI #

The optional Hiveloom workbench is a chat-first UI for creating, testing, running, inspecting, and improving harnesses. It ships separately — as an npm package rather than a Python one — so production environments only carry the framework:

npx hiveloom-workbench
# http://127.0.0.1:8770

Nothing to install: npx fetches the interface, its Python API, and the launcher that wires them together, then runs them against whichever interpreter already has hiveloom. hiveloom ui is a shortcut to the same command. Contributors can run the UI and API with hot reload from a source checkout via devtools/ui/dev.sh.

API keys #

hiveloom talks to whichever lab the harness names. The default provider is claude, which reads ANTHROPIC_API_KEY:

export ANTHROPIC_API_KEY=sk-...

hiveloom models lists every builtin provider and the environment variable it expects — OpenAI, Gemini, Mistral, DeepSeek, xAI, Groq, OpenRouter, Together, Fireworks, Ollama, and vLLM are all builtin. See Models and providers for switching an existing harness.

Local providers (Ollama, vLLM) need no key, which makes them a reasonable way to try the runtime before spending anything.

From source #

git clone https://github.com/FrancescoMrn/hiveloom.git
cd hiveloom
uv sync --extra dev

Agent guidance #

Guidance for coding agents ships inside the wheel — no network fetch required:

hiveloom guide --list