Ananya Roy

One harness for every coding agent: swap models and runtimes without rewriting

You want your team using the best coding agent for each job : Claude Code, Codex, Cursor, or one you built yourself , without rewiring your setup every time a better model ships, and without each agent running loose with its own credentials, spend, and idea of what's safe.

Across the thousands of engineering teams we work with globally, the pattern is the same: every agent is its own silo, so you either write glue code for each vendor or lock yourself into one. This session shows a bran new approach : Omnigent , a meta-harness layer that treats the runtime, model, tools, prompts, and safety policies as separate, swappable parts. With Omnigent, you can switch its model or its underlying harness without changing anything else.

We'll walk through how it works. A single governance layer sits over every agent and checks each action: tool call, model request, file write, with an allow, ask, or deny decision. The same layer routes models to manage cost, caps spend, and gates risky actions, while sessions run sandboxed and can be shared live for teammates to review and steer from any device.

You'll leave knowing how the meta-harness layers your own agents so you're never tied to one vendor, how to govern spend and risk with policies, and how your whole team can collaborate in real time.