ivankuznetsov/hive

<p align="center"> <img src="docs/assets/logo.svg" alt="Hive" width="96" height="96"> </p>

Hive

Website & docs: hivecli.sh — quickstart, concepts, the full command reference, and an llms.txt for agents.

Community: join the Hive Discord for questions, feedback, and release discussions.

Hive turns a rough software idea into a merge-ready pull request through a multi-agent pipeline you can watch as it works. You sketch an idea in a few sentences, open the hive tui dashboard, and watch the work move forward: brainstorm pins down what you actually want, plan fixes the approach, execute writes the code, review hardens it, and finalize ships the PR. You can step in at any stage with a normal editor — every artefact is a markdown file in a stage folder, inspectable and editable by you or by another agent.

The mental model is folders. Every task is a directory; the folder's location is the task state. Moving a task from 2-brainstorm/ to 3-plan/ is the approval gesture, and every stage writes a durable artefact the next stage can trust. That practice — making each step's output strong enough for the next one to run autonomously — is called compound engineering. It's how Hive carries work from rough idea to merged PR while letting humans drop in on their own terms instead of a chat thread's.

1-inbox  ->  2-brainstorm  ->  3-plan  ->  4-execute  ->  5-open-pr  ->  6-review  ->  7-artifacts  ->  8-finalize  ->  9-done
capture       refine           design      build          draft PR       harden        collect        publish         archive

Read the model in depth in docs/concepts.md — folder-as-agent, the marker protocol that lets stages negotiate handoff, the nine stages in detail, and the trade-offs that come with making everything a file.

Demo

A sub-two-minute reel of the whole loop — install, hive init, capture an idea in t