Introducing OmniConflux: fleet orchestration and cross-model verification
Software engineering is shifting from chat assistants to agents that work in parallel. OmniConflux is designed to pool the machines you own into a fleet and to have a verifier from a different vendor check each finished task.
Beyond single-agent chatbots
The first era of AI software development was single-session conversations: you typed a prompt, inspected the code, copied it into your editor and ran the tests by hand. That works for small snippets. It strains when a task spans a multi-file refactor, a database migration or a long test run. Teams need agents that work in parallel, split a large feature into distinct tasks, run builds and hand back work for review.
The silent failure of self-verification
When the model that wrote a change is also the model that reviews it, the review shares the author's blind spots. A mistake the author did not see is a mistake the same reviewer is likely to miss. That is how regressions slip through and how human review time gets spent on work that should have been caught earlier. We do not think an agent should grade its own homework.
The rival verifier
OmniConflux is designed around a cross-vendor check. Each finished task from a specialist agent is designed to go to a verifier from a different vendor, together with static checks and sandbox test runs that need no model at all. In the design, the verifier reads the task, the diff and the recorded evidence, then returns approve or rework with reasons. For changes you mark as high impact, the design asks for more than one independent reviewer. Verification adds a check with evidence. It is not a promise that every result is correct.
Your machines are the cloud
Running many agents in parallel takes real compute. You may already own some of it: a desktop, a laptop, a server in a closet. OmniConflux is designed around that. It is designed to pool your existing machines into one fleet, so heavy builds and test runs can run on hardware you own while you steer from a laptop.
Voice, by design
Managing several agents through terminal tabs gets messy. Voice is designed as a hands-free way in: the design transcribes your speech on your machine and sends the text as a command to the agent you chose. Voice is not released, and the controls will be documented at launch.
What comes next
OmniConflux is being built for macOS and Linux. Availability will be announced through the closed-beta pre-registration page. Aura by OmniConflux, a planned ad network, is also on the roadmap. It is not live.