Sovereignty · 2026-09-16 · 1:38

A task board where AI agents are team members

Name: A task board where AI agents are team members
Uploaded: 2026-09-16
Duration: 1 min 38 s
Description: The interesting idea is not the board but the model: agents with scoped permissions and assigned tasks. Worth studying before you wire your own agents together.

itsaplan is a self-hosted project tracker (a Linear/Jira alternative) built around AI agents as first-class members: they hold permissions and take assigned work, triggered by mention, assignment or schedule.

In favour

Self-hostable with Postgres; external agents (Claude Code, Copilot CLI) connect over a simple runner.
Per-agent permissions make least-privilege the default, not an afterthought.

Worth watching

AGPL-3.0 core: fine to self-host, but offering it as a service to clients triggers source-disclosure.
Pre-stable — the authors warn of breaking changes before the first release.

Our takeThe interesting idea is not the board but the model: agents with scoped permissions and assigned tasks. Worth studying before you wire your own agents together.

Source: itsaplan (croffasia, GitHub)

← All Focus posts

AI-narrated · AI-generated content

Transcript

What if the AI agents on your project were not tools you call, but members of the team, on the same task board as everyone else? One self-hosted project tracker is built around exactly that. It is an alternative to the familiar trackers, and in it, AI agents are first-class members. They hold permissions, and they take assigned work. An agent starts when someone mentions it, when a task is assigned to it, or on a schedule. You can host it yourself, backed by a standard database. And external coding agents, the ones that run outside the tracker, connect to it over a simple runner. Permissions are set per agent. That makes least privilege the default, not an afterthought. Each agent can touch only what it has been given. Two things to keep in mind. The core is licensed AGPL-3.0. Self-hosting it is fine. But offering it as a service to your clients triggers source disclosure. And the project is pre-stable. Its authors warn of breaking changes before the first release. So what you set up today may need reworking tomorrow. So the interesting idea is not the board. It is the model: agents with scoped permissions and assigned tasks. That is worth studying before you wire your own agents together.