Agentum

Ship while you sleep.

Agentum is the ADE. Design and run autonomous engineering pipelines on infrastructure you own. A spec goes in. A PR comes out.

Download for macOS → Windows, macOS & Linux
The Agentum desktop app — workspaces, sessions, and a live agent terminal.
What developers build with it
PM agent Architect agent Developer loop Reviewer agent Verify gate → PR merged ✓
How it works

Design the pipeline. Let it run.

Agentum gives you the building blocks to assemble an autonomous engineering pipeline. You define how the phases connect — Agentum runs them, watches them, and keeps them going on your infrastructure.

01
PM agent
Turns your vague ticket into checkable acceptance criteria and a concrete scope. Refuses to pass garbage downstream.
→
02
Architect agent
Validates the design, picks the right files to touch, and writes it all down before a single line of code is written.
→
03
Developer loop
Implements each item one at a time. Every item goes through a verify gate before the loop advances. No skips.
→
04
Reviewer agent
Reads the full diff, raises concerns, and signs off — or kicks it back to the developer loop with a reason.
→
✓
PR merged
Every gate green. Human review is opt-in, never the default. You come back to shipped code.
The loop

Spec. Build. Review. Ship.

Agentum runs the full engineering cycle on a host you own. Each phase is autonomous — no human relay between spec and PR. Close the lid and come back to merged code.

01 — Harness

Every major coding agent, one harness.

Claude, Codex, Gemini, Cursor, Hermes, OpenCode, Aider — Agentum probes what's installed and assigns each phase to the right tool. The harness handles orchestration so no agent needs to know about the others.

agents · installed
Claudeready
Codexready
Geminiready
Cursorready
Hermesnot on PATH
OpenCodeready
02 — Task queue

Atomic task handoff. No collisions.

An atomic-claim kanban every agent in the pipeline can see. One phase finishes and drops the card; the next claims it — first claim wins, the rest get a 409. No two agents ever work the same task.

board · atomic claim
Todo 2
auth refactor
write docs
In progress 2
api endpointscodex
settings UIclaude
Done 2
test suite
lint pass
409first claim wins — the rest pick another card
03 — Deploy targets

Run the ADE anywhere. Switch in a keystroke.

Save your local machine, a VPS, or a homelab as profiles — each with its own TLS trust-on-first-use and token. The ADE runs wherever you point it; swap targets from the TUI (Ctrl-S) or the app.

profiles
local127.0.0.1:8787active
vpsmy-vps.example.com:8822TLS pinned
studio10.0.0.4:8787LAN
CtrlSswitch endpoint
Mission control

Watch the autonomous loop, live.

Every pipeline phase across your local machine and remote hosts in one view — running, waiting, crashed. See exactly where your ADE is in the spec → build → review cycle without touching a terminal.

See the architecture →
agentum — 2 hosts · 5 sessions
2 running
local macOS · darwin-arm64
3 sessions · 1 running
alpha claude · opus-4.8ctx 78%running
review-pr codex6 filesneeds input
docs-pass geminiidle 14midle
vps linux · ssh
2 sessions · 1 running
cron-nightly hermesdeps + testsrunning
build-arm claudewatchdog · session.crashedcrashed
Install

Two ways to run it.

Start with the desktop app — the full agent environment. Keyboard-first? It ships a daemon, a fast TUI, and a complete CLI too.

Agentum

The agent environment, native.

Workspaces, the board, channels, and live agent terminals — one app on Windows, macOS & Linux. The fastest way in.

Download for macOS → Windows, macOS & Linux · x64 & arm64
Get started

Your ADE.
Your infrastructure.

One install. Point it at any host you own. Build your autonomous development pipeline and let it run.

Download for macOS → View source
MIT · single binary · Windows, macOS & Linux · no cloud