Skip to content

Practical fleet workflows

Real jobs for your agent fleet

Six machines, phone to server rack, all connected to one glowing hub

OmniConflux is designed to pool your own machines into one agent fleet. These are the jobs it is designed for: migrations, code reviews, social posting, voice, computer use and private security testing. Every agent task gets a verifier from a different vendor.

Codebase migrations

Designed to refactor legacy code, upgrade major dependencies and adapt to breaking API changes on separate branches, without blocking feature work.

Explore platform architecture

Outcome: A migration branch ready for your review.

Checked by: a verifier from a different vendor, before the task counts as done.

Code and security audits

Designed to review code and proposed changes for bugs and security issues in sensitive repositories before pull requests open.

See verification engine details
A glass prism taking in one white beam and sending out three checked beams

Outcome: Findings with the evidence a reviewer needs to confirm them.

Checked by: a verifier from a different vendor reviewing the findings.

One logo going into a press and coming out as correctly sized frames in 1:1, 4:5, 16:9 and 9:16

Native social publishing

Designed to turn engineering updates into posts written for each social network, in your own brand voice.

Outcome:
Draft posts shaped for each network, ready for your review.
Checked by:
a verifier from a different vendor. Publishing approval controls will be documented at launch.
Explore social media management
A studio microphone with a glowing waveform ribbon flowing out of it

Voice-driven task orchestration

Designed to let you speak a goal and send it to an agent on one of your machines without reaching for the keyboard. Speech is designed to be transcribed locally. Voice is planned.

Outcome:

A spoken goal is designed to become a command for one of your harnesses.

Checked by: a verifier from a different vendor. Phone approvals are planned.

Read computer use and voice story

Computer use and OS automation

Designed to operate desktop applications, step through staging sites, and check interface states on machines you own. Computer use is planned, and detailed controls will be documented at launch.

Outcome:
Repetitive desktop steps done by an agent, with each action designed to be shown to you.
Checked by:
a verifier from a different vendor reviewing the finished task. Approve once, for a session, or always is planned.
View computer use overview
An open laptop in dark glass, its screen lit violet

Private security testing

Closed beta: authorized, defensive testing of systems you own, with frontier-class security AI, privately hosted. Requests for access are open.

Outcome: A record of what was tested and what was found, on systems you are authorized to test.

Checked by: a verifier from a different vendor and your authorized team.

Review private security AI

Digital Sovereignty

Open by design: any model, your hardware

OmniConflux is designed never to lock you into a single vendor or proprietary cloud silo. It is designed for your own API keys, local open-weight models on your private hardware and the command-line tools you already use. You control the hardware and the code repositories.

Frequently asked questions

Why does a different vendor check the work?

An agent that marks its own work tends to agree with itself. Every agent task gets a verifier from a different vendor, so a second perspective reviews the result before it counts as done. It is a check, not a guarantee.

Can these workflows run completely on local models?

That is the design. OmniConflux is designed to work with local open-weight models on your own machines, so a workflow can run without sending code or prompts to a cloud endpoint. OmniConflux 0.3.0 for Omarchy, macOS on Apple silicon and Windows is in closed beta. Approved accounts download the app inside their account.

How do human operators retain control over automated agents?

Human approval is part of the design. The plan is to approve once, for this session, or always, from your phone. Which actions ask for approval will be documented at launch.