Ota: The execution contract for AI agents

AI agents increasingly work across software repositories, but the truth about how a repository should be prepared, verified and executed is often fragmented across documentation, scripts, containers, CI workflows and operational policy. An agent may follow a plausible command while using the wrong environment, dependencies or authority boundary.

Ota addresses this problem through a machine-readable execution contract shared by developers, CI and AI agents. A repository can explicitly declare:

• prerequisites and readiness checks;

• permitted agent tasks and protected paths;

• native, container and other execution modes;

• task dependencies and required services;

• verification requirements;

• execution evidence and the limits of what each result proves.

Ota is a working open-source product implemented in Rust. It has been tested across real repositories, Linux, macOS, Windows, containers and CI environments. It does not claim that a passing check proves application correctness, production approval or general AI safety. Its purpose is to make a specific execution boundary explicit, enforceable and reviewable.

For the UK–Singapore CR&D programme, Spydar AI proposes using Ota as the technical foundation for joint industrial research into execution assurance for AI agents in regulated software environments. Potential research areas include bounded agent authority, independently verifiable execution evidence, enterprise-policy integration, adversarial evaluation and cross-organisational assurance workflows.

We are seeking a Singapore R&D partner with complementary expertise, engineering capacity, relevant experimental environments and a credible route to commercialisation. The resulting project must involve genuine technical co-development rather than routine product deployment or resale.

Website: https://ota.run
Open-source repository: https://github.com/ota-run/ota