Strust CLI · Developer preview

Install the agent that checks the change.

Strust combines a Codex-style coding agent with an independent behavioral evaluator. Work in the terminal, then replay the old and new software on the same inputs before the change reaches review.

See how verification works

Native packages by operating system · Built on open-source Codex 0.144.1

Choose your platform

Strust highlights your detected operating system without starting a download. Every platform remains visible.

Native builds · one evaluator contract

Coming soon

Windows

Windows x64

Portable preview with the evaluator runtime included.

Coming soon

macOS

Apple silicon + Intel

Signed universal package planned after native evaluator testing.

Coming soon

Linux

Linux x64

Native tarball planned after clean-machine evaluator testing.

First run

From ZIP to the TUI in three commands.

The package includes the agent, evaluator kernel, embedded runtime, offline demo, licenses, and release evidence. It does not install a service or write to the registry.

Windows instructions

PowerShell

Expand-Archive .\strust-0.9.0-rc.1-windows-x86_64.zip .\strust
cd .\strust\strust-0.9.0-rc.1-windows-portable
.\strust.exe

On first launch, choose ChatGPT sign-in, an API key, or another configured model provider.

One download, two judgments

The agent can change the code. It cannot certify itself.

Agentic terminal

An interactive coding-agent TUI, headless exec mode, provider configuration, MCP, skills, and sandbox controls.

Independent evaluator

Capture baselines, replay the reference and candidate, minimize counterexamples, and locate the first divergence.

Reviewable evidence

Keep findings, receipts, run history, and exact replay commands outside the agent's own judgment.

Strust runs commands declared by the project on your machine. Use code you trust; a copied workspace is not a hostile-code sandbox.