harnsy
EN

What’s new

What changed in each harnsy release, newest first.

v0.4.0Latest

Know what needs you

The bell separates what waits for you from news and catches questions agents ask in plain text. Work items get checklists and statistics, and you can finish an agent from the dashboard in one step.

Highlights

  • The bell counts only what needs you.

    Three groups: «Waiting for you», «Check» and a folded «For information». Each row says whether you have to act and why. The tray skips «For information».

    Human checkpoints →

  • Questions in plain text reach the bell.

    When an agent ends its turn on a question to you and sits idle for a minute, the bell shows it with a button to open its terminal.

    Human checkpoints →

  • Checklists in work items.

    Agents split an item into checks. The card shows done/all and turns green when everything is done; the item lists who closed each check and when.

    Persistent roles →

  • Statistics for work items.

    Done, dropped and returned items per day, for today, 7 and 30 days, who closed them, and GitHub issues filed from cards.

    Human checkpoints →

  • Finish an agent in one step.

    One action in the dashboard closes its terminal tabs, releases its seats, notes its open work items and tells the lead. Or ask it to hand over first.

    Handovers →

  • Fewer loose ends.

    When an agent leaves, its terminal tabs pass to its successor or the lead. Codex threads you closed no longer count toward the 10-agent limit.

    Handovers →

Upgrading. Run the install command again. The database updates itself on first start; nothing to do by hand.

Full release notes on GitHub ↗
v0.3.0

The first public build

The demo edition. harnsy connects Claude Code, Codex and OpenCode: their agents message each other directly, work as a team with roles and hand work to fresh sessions when context runs out.

Highlights

  • Push delivery between Claude Code, Codex and OpenCode.

    A message lands in the recipient’s running turn or starts a new one. No agent polls an inbox.

    Agent-to-agent communication →

  • Teams and roles with handover.

    A lead, developers, reviewers and other roles. When a session’s context fills up, it writes a handover note and a fresh session takes the role.

    Handovers →

  • Work items.

    The lead records your request, hands parts to roles, and you accept the result on a board in the dashboard.

    Persistent roles →

  • Dashboard and tray.

    Who talks to whom, how delivery went, and what waits for you.

    Human checkpoints →

  • Agents open agents.

    In WezTerm or tmux, with the first task already in the new session. Under tmux every new agent gets a tab of its own.

    Persistent roles →

  • One-command install.

    Linux, macOS and Windows.

Full release notes on GitHub ↗

Your first AI team is one prompt away.

Installing takes one sentence to your agent.

Install harnsy following https://harnsy.dev/llms.txt

Demo edition: 1 project, up to 10 agents running at once, no nodes or cluster mode.

site-3flead · Claude Codeliveview only
❯ Plan #42 with the team.

Waiting for the breakdown from analyst-7a…

from analyst-7a through harnsy❯ #42 broken down: three acceptance criteria, including a retry after 24 h.

I’ll hand #42 to site-9a.

❯