The personal AI operating system
The operating system for an AI workforce.
amoOS plans the work, dispatches it to Claude and Codex agents across a fleet of machines, verifies what comes back, and compiles everything it learns into a Brain. It gets better with every run.
951
Tasks completed
Across every active project
342
Pull requests merged
Verified before merge
9.2B
Tokens in 30 days
Across 12 models
98%
Prompt-cache hit rate
Context reused, not re-paid
More than a dispatcher
amoOS isn’t a wrapper around a coding agent. It is the layer above them: it decides what to work on, where to run it, how to check it, and what to remember.
A fleet, not a laptop
amonode
Linux · always on
60%
amomini
Mac mini
18%
amom4pro
Cockpit
22%
A Brain that compounds
Evidence before done
Reachable from anywhere
The operating loop
01
Plan
Goals become sprints with executable, dependency-aware tasks.
02
Dispatch
Each task goes to the right model on the right machine.
03
Observe
Live sessions, capacity and questions stream into one place.
04
Verify
Work is accepted on evidence, not on an agent saying “done”.
05
Learn
Reports and lessons compile into the Brain for the next run.
The command center
Sprint progress, live sessions on every machine, capacity, interventions and verification signals in one view. Longer autonomous runs are only useful if they stay legible.
Command center
3 nodes onlineSprint
18 / 24
tasks done
Fleet
7 live
sessions
Capacity
62%
weekly used
brain-recall-eval
amonode · Opus 5.5
gateway-rate-limits
amomini · Codex
voice-latency-pass
amonode · Sonnet 5.5
digest-composition
amomini · Opus 5.5
usage-pacing-v2
amonode · Codex
Merged this sprint
57
Illustrative data
Human in the loop
Step away without disappearing.
Safe, reversible work keeps moving. When an agent hits a real judgment call it asks instead of guessing, and the question lands on your phone with options. One tap and the worker resumes.
- 134 decisions answered from the phone in the last 30 days.
- Destructive actions always ask first; nothing irreversible is guessed.
- Unanswered questions fail safe: gated actions are denied on timeout, never assumed.
amoOS
online · your fleet
FAQ
What is amoOS?
amoOS is a personal AI operating system built and run by Aung Myint Oo. It plans work as sprints, dispatches tasks to Claude Code and Codex agents running on a fleet of machines, verifies the results, and compiles everything it learns into a persistent Brain that makes the next run faster.
Is amoOS just a coding-agent dispatcher?
No. Dispatch is one part. amoOS also includes sprint planning, a verification gate, a compiled knowledge Brain with provenance, learning loops that attach lessons to future tasks, morning and evening briefs, content and relationship tools, usage telemetry, and four ways to reach it: Telegram, a macOS voice app, a web command center and MCP.
Which AI models does amoOS use?
amoOS is model-agnostic. It routes work by difficulty: Claude Opus 5.5 for coordination, hard tasks and review, Claude Sonnet 5.5 for standard and mechanical work, Codex (GPT models) for bulk execution on a separate subscription pool, and small or local models for utility jobs. Twelve models ran in September 2026.
Does amoOS keep working when the laptop is closed?
Yes. Agents run on always-on machines, a Linux box and a Mac mini, each in an isolated git worktree. The laptop is only the cockpit for planning and review, and decisions can be answered from a phone through Telegram.
How does amoOS get better over time?
Every session ends with a bounded handoff that records what happened, which lessons helped, and new lesson candidates. Reviewed lessons are attached automatically to future tasks, and project knowledge is compiled into canonical Brain pages that every agent reads before it works.
How does amoOS make sure agent work is correct?
Work is accepted on evidence, not on an agent saying it is done. Definition-of-done checks and independent verify passes run before merge, genuine judgment calls are escalated to a human, and destructive actions always ask first.
How much work does amoOS actually do?
Between 27 July and 29 September 2026 amoOS ran 920 agent sessions, completed 951 tasks and merged 342 pull requests across three machines. In the last 30 days of that window it processed 9.2 billion tokens at a 98% prompt-cache hit rate.
Can I use amoOS?
amoOS is a personal system that Aung Myint Oo builds and runs every day, not a commercial product. Follow @amodev on X for updates on how it works and what it ships.