It runs on your machine
One small harness on 127.0.0.1 owns every agent process. Transcripts, keys and events live in ~/.pragma, not in somebody's cloud.
A desktop app where every chat is a real AI agent, running on your own machine.
Live, not a screenshot. Tap Allow when Otto asks.
One small harness on 127.0.0.1 owns every agent process. Transcripts, keys and events live in ~/.pragma, not in somebody's cloud.
Bots run on the claude, codex, qwen and grok CLIs installed on your own computer. Your logins, your subscriptions, no proxy in the middle.
Every shell command, file write and question arrives as a card in the thread. Allow it, deny it, or just answer back.
In motion
01 · The roster
Work isn't one conversation. Pick one and watch it.
Grab one. Throw it.
02 · Brains
Switch a bot's brain mid-sentence. The thread doesn't reset.
Changes the badge in the window above, live.
03 · Hands
Its own machine. Watch it work, or take the controls.
04 · Approvals
Shell, files, spending, sending. Each one is a decision you make.
~/work/pragmapnpm test --filter server
05 · Reach
Connect an account once. Every bot can use it.
and 500+ more through Composio
06 · Local first
A 9B running on your own GPU has no computer, no browser, no skills and no way to draw. Pragma hands it all four, so the gap between it and a frontier model stops being about what it is allowed to touch.
~/.pragma
harness listening on 127.0.0.1
07 · What it costs
No account, no waitlist, no card. Pragma adds no model bill of its own: it runs on the Claude, Codex or Qwen subscription you already pay for, or on a model running on your own machine, where it costs nothing at all.
08 · The claim
Not a chat window with a model behind it. A roster of agents that hold their own threads, run on your machine, and ask before they act.
Version 1.3.14, for Windows, macOS and Linux.
Free while we launch. Bring your own agent: Pragma runs on the Claude, Codex or Qwen subscription you already pay for, and adds none of its own.