TEREO

AI writes. TEREO proves.

TEREO is a proof layer for AI coding. keep only if gain > noise Each kept receipt becomes the next baseline.

It watches one small change, reruns one fixed check, and writes one readable receipt. The check should show one gain and catch the core breakage that would make that gain false. The check can prove a bug fix, a behavior change, or a speedup.

TEREO comes from Ancient Greek τηρέω (tēréō): to watch over, guard, preserve.

Want the full story? Read the blog post: What Is TEREO?

One loop.

One kept receipt.

Demo

1. Install TEREO.

pip install tereo

2. Feel the loop.

tereo demo

If you prefer module execution, python3 -m tereo demo is the same runtime.

What it looks like:

Working on this repo itself? See the local setup in runtime/README.md.

The demo should close with something like this:

scorecard: latency: 10.000000 ms -> 8.000000 ms; 2.000000 ms better (20.00%)
verdict: keep
confidence: high

Quick Start

1. Install TEREO.

pip install tereo

2. Run TEREO locally.

tereo --help

If you prefer module execution, python3 -m tereo --help is the same runtime.

Use your own check. Keep your own workflow. Make it yours.

TEREO is small on purpose. Fork it. Keep it. Make it yours.

Public Commands

tereo demo
tereo prove
tereo show
tereo log

Everything else is there to support this surface, not to replace it.

Why This Is Hard

• one good-looking run can lie
• gains drift
• noise changes verdicts
• checks matter more than vibes
• hold is different from keep

Narrow By Design

TEREO is strongest when one fixed check can judge one small change. That check should include both what should get better and what must stay true. If a patch hits the promise but breaks the core, that is a false gain, not a keep. That is a feature, not a full replacement for broader product judgment.

Read Next

• START.md Start here.
• runtime/README.md The runtime, advanced commands, receipts, and proofboard.
• skill/README.md The portable rulebook for humans and agents.
• WHY.md Why TEREO stays this small.

If the project is easy to explain, it is easy to trust.