Install

hiveloom 1.3.0 requires Python 3.11 or newer and is available on PyPI.

Install the CLI or Python library #

For an isolated CLI, install or upgrade with:

uv tool install --upgrade 'hiveloom==1.3.0'
hiveloom --version

For a Python project, add it as a dependency instead:

uv add 'hiveloom==1.3.0'
# Alternatively:
# pip install 'hiveloom==1.3.0'

Follow the Quickstart to build, validate, and dry-run a harness. Schema/catalog inspection and dry-run make no model call; declared MCP servers are still contacted for tool discovery during dry-run.

Workbench #

The workbench ships separately as hiveloom-workbench on npm. The new Research, provider settings, harness settings, and fork-resume features require the 0.2.0 workbench and hiveloom 1.3.0. At this documentation release, npm still offers 0.1.0; npx hiveloom-workbench and hiveloom ui fetch that older release.

Use the tagged source for the new UI now. You need Node.js 20 or newer and uv:

git clone --branch v1.3.0 https://github.com/FrancescoMrn/hiveloom.git
cd hiveloom
uv sync --extra dev
devtools/ui/dev.sh
# http://127.0.0.1:5173

The source launcher installs frontend dependencies on first launch. Open Settings → Models to add a provider and choose the default model. See Workbench for provider storage, harness settings, and fallback.

Once workbench 0.2.0 is published, its packaged launcher can be used instead at http://127.0.0.1:8770. Check the npm package for availability.

Develop from the release source #

The same tagged checkout provides the runtime and its offline examples:

uv run hiveloom --version
uv run hiveloom schema --json
uv run hiveloom catalog tools --json

Use uv run hiveloom for commands inside this checkout. Older 1.2.0 installation instructions remain available.

Provider credentials #

The default executor provider is claude, which uses ANTHROPIC_API_KEY:

export ANTHROPIC_API_KEY=sk-...
hiveloom models --json

CLI runs read credentials from the process environment or the harness's own .env. Workbench-saved credentials are local to that workbench setup; they are not automatically exported to a separate CLI process. Local providers such as Ollama and vLLM do not require a provider key, but do require a running server. Scripted lab fixtures need neither credentials nor a server.

Guidance for coding agents #

The wheel includes the schema, catalog, reference guides, and focused skills:

hiveloom guide --list --json
hiveloom guide research --json

See Models and providers to select an executor, a proposing model, or a research director. These are separate choices with separate costs.