How Grex runs a mission
Hand a bounded task to AI agents on computers you own. A second agent reviews the exact result against your acceptance criteria before anything is accepted.
- 1
Write the mission.
Give the objective, the acceptance criteria, the files it may change, a starting commit and a deadline.
- 2
A builder does the work.
A builder agent works in a sandbox on one of your own machines.
- 3
Grex checks it.
Grex runs mechanical checks, such as the build and tests, and keeps the patch, the logs and the check results.
- 4
A second agent reviews it.
A separate reviewer agent judges the exact submission against each acceptance criterion. We call this rule Duoviri: the agent that did the work never accepts it.
- 5
You decide.
You see the result and the reviewer's verdict on each criterion. Accepted code becomes a pull request. A rejected mission can be corrected with the reviewer's reasons attached.
- 6
Run it again on a schedule.
Research briefs and site reports can run daily. A schedule pauses itself after repeated failures.
Where it runs
A small control plane stores missions, sends work to your machines and keeps the evidence. It runs no AI.
Your machines, called Nidi, do the work. Today these are Linux and Windows computers. Each one can run several missions at once.
Grex uses the agent command-line tools you install and sign in to yourself: Codex with a ChatGPT plan, and Claude Code with a Claude plan. It can also call the MiniMax, Anthropic and Gemini APIs, which are billed per use, or a local model through Ollama. The builder and reviewer are chosen separately.