mellona-hive 2.3.0

CLI-first orchestration platform for autonomous agents with Git-backed task, run, and memory state

Agent Hive

Agent Hive is a repo-native control plane for autonomous work. Use Codex, Claude Code, or local/manual execution to do the work. Use Hive to supervise tasks, runs, memory, policy, and approvals from one place.

Keep your agent. Add a control plane.

mellona-hive is the distribution you install from PyPI or Homebrew. Mellona is the package family. Agent Hive is the current product. The base install gives you the hive command. Add mellona-hive[mcp] when you want the thin hive-mcp adapter.

Install

Pick the installer you already trust:

uv tool install mellona-hive

pipx install mellona-hive

python -m pip install mellona-hive

brew tap intertwine/tap
brew install intertwine/tap/mellona-hive

Then verify:

hive --version
hive doctor

The full install matrix, lane chooser, and maintainer split live in docs/START_HERE.md.

If you want optional extras:

• Observe console: uv tool install --upgrade 'mellona-hive[console]'
• MCP adapter: uv tool install --upgrade 'mellona-hive[mcp]'
• Hosted E2B sandbox: uv tool install --upgrade 'mellona-hive[sandbox-e2b]'
• Self-hosted Daytona sandbox: uv tool install --upgrade 'mellona-hive[sandbox-daytona]'
• Homebrew currently ships the base CLI only, so add extras through uv tool, pipx, or pip

If you install sandbox extras, verify the current machine with:

hive sandbox doctor --json

If you are testing before the first tagged public release lands on PyPI and Homebrew, use the git install:

uv tool install --from git+https://github.com/intertwine/hive-orchestrator.git mellona-hive

First Run

Start in an empty directory:

mkdir my-hive
cd my-hive
git init
hive onboard demo --prompt "Create a small React website about bees."

That gives you a real workspace, a starter project, a conservative PROGRAM.md, and a first task chain with one ready task. If you want the observe-and-steer console, install mellona-hive[console] first, then run hive console serve.

Fresh onboarded projects may start with the placeholder local-smoke evaluator so the loop works immediately. Replace it with a real repo-specific evaluator before you trust autonomous promotion.

If the first hive finish later says there was nothing to promote, that is usually a healthy noop rather than a broken setup. To intentionally see a successful first promotion, make one tiny docs-only change while working the demo task, then finish the run.

Then use the normal loop:

hive next --project-id demo
hive work <task-id> --owner <your-name>
hive finish <run-id>

Optional extras stay intentionally small:

• mellona-hive[console] adds the observe-and-steer console
• mellona-hive[mcp] adds the thin search + bounded local execute adapter

Choose The Right Guide

• Start here: docs/START_HERE.md
• Fresh workspace walkthrough: docs/QUICKSTART.md
• Existing repository adoption: docs/ADOPT_EXISTING_REPO.md
• Maintainer and release docs: docs/MAINTAINING.md and docs/RELEASING.md