Show HN: Hive – A full dev workspace (kanban/session modes+multi-repo+agent sdk)

Show HN: Hive – A full dev workspace (kanban/session modes+multi-repo+agent sdk)(hive-ai.dev)

1 points by moropex 41 days ago | 2 comments

I kept hitting the same problem with Claude:

The native Claude app is great but it can be much better when you unlock capabilities of desktop rather than the terminal. Such as:

- no task management

- no structure

- hard to work across multiple repos

- everything becomes messy fast

So I built a desktop app to fix that.

Instead of chat, it works more like a dev workspace:

• Kanban board → manage tasks and send them directly to agents

• Session view → the terminal equivalent of Claude code for quick iteration when needed/long ongoing conversations etc

• Multi-repo “connections” → agents can work across projects at the same time with context and edit capabilities on all of them in a transparent way

• Full git/worktree isolation → no fear of breaking stuff

The big difference:

You’re not “chatting with Claude” anymore — you’re actually managing work.

We’ve been using this internally and it completely changed how we use AI for dev.

Would love feedback / thoughts

It’s open source + free

GitHub: https://github.com/morapelker/hive

Website: https://morapelker.github.io/hive

Gingiris1031 41 days ago |

The kanban-as-task-queue framing for AI agents is smart — it forces you to decompose work before handing it off, which tends to produce much better outputs than free-form chat. One challenge we ran into with multi-repo agent workflows is context window management when files across repos have conflicting conventions. Does Hive do anything to normalize context before feeding it to the agent?

moropex 41 days ago | |

so the way hive handles cross repos, is by creating a temporary directory, with symlinks to the original worktrees. For all the model cares and knows, hes working on a mono repo of x projects, and is able to navigate it the way it knows best. All edits he does - are then propagated to the original worktrees in their original folders. So you kind of get the best of both worlds (without requiring you to feed the context manually)