Calibrated AI you can audit

The shared brain for
your AI fleet.

Your AI practices keep an auditable, calibrated record of what they learn and do, share only what matters with each other, and never act on shared work without your sign-off.

Book a demo See how it works

Or grab the free honesty prompt for any chat.

Built on open source: empirica (MIT) and ecodex (Apache-2.0).

# before any change, a practice has to pass the gate
empirica preflight   # declare what you are about to do
empirica check       # the Sentinel: proceed only if grounded
empirica postflight  # measure belief against real evidence

The difference isn't the model

The same underlying intelligence either scrambles up the cliff with no gear, or climbs it equipped, roped, and tracked. Empirica is the scaffolding.

Split image: on the left a lone climber scrambles up a cliff with no gear, map, or rope, guessing at every step (raw AI); on the right an Empirica-equipped climber ascends with helmet, rope, and a guide, supported by structured knowledge, calibration, and memory.

Not another agent. A practice with a track record.

An agent is a model in a loop, with no real memory and no way to prove what it did. A practice is different. It remembers, it shows its work, and its confidence is checked against reality.

📜

A full history, from day one

Every practice keeps a record of what it has learned and done since the moment it launched. You can look back and see exactly what it did at any point in the work, and why.

🎯

Shares what matters, not everything

A practice passes on the few details that count, grounded in what it has actually learned, to the practitioners who need them. No memory dump, no noise, just the knowledge that moves the work forward.

🎚️

Measured, so you can trust it

Every step is checked against real evidence, so a practice's confidence is earned, not claimed. The result is work that is auditable and reliable, not a black box that says "done" and hopes.

🔓

Open foundation, principles you can inspect

The discipline that makes this work is an open-source framework. The rules every practice follows are transparent and yours to read, audit, and run yourself. Trust is built on principles you can see.

What your fleet gains

One practice is reliable on its own. The advantage compounds when many work together.

🧠

Knowledge that compounds

What one practice learns becomes available to the others, grounded and on demand. Your team's hard-won context stops living in one person's head or one chat window.

What it looks like: a developer's AI logs a security gotcha in one project. The next day, an AI in a different repo hits a related path and is handed that exact lesson, with no one re-explaining.

🛂

Nothing acts without your sign-off

When one practice wants to act on another's work, the request comes to you first. A real decision point built into how they coordinate, not a confirmation box bolted on after the fact.

What it looks like: a practice proposes a change. It lands on your phone or in the browser extension. You accept or decline. Then, and only then, it ships.

🤝

They agree before they act

Practices coordinate the way a careful team does: gather the facts, agree on the plan, then act. The agreed plan is recorded, so it survives long after the conversation that produced it.

What it looks like: two practices in different projects compare notes, settle on an approach, and bring you one clear proposal instead of two half-formed ones.

This site is a working example. Its shape was settled by eight practices comparing notes and correcting each other before a single page was written. See how it works →

Run it across your whole fleet

Cortex connects your practices into one shared, coordinated layer (the mesh). The extension is where you watch it and govern it. Together they are the product, for a single developer through a whole team.

🧠

Cortex

early access

The shared-brain backend. It holds the calibrated knowledge your practices build, routes their requests through your approval, keeps shared plans that outlast any session, and isolates every organization's data on its own.

🔭

The extension

beta

The control plane in your browser. See what every project and practice is doing, and accept, decline, or refine the actions they propose. The single place you run the whole fleet, without opening eight terminals.

The extension: the fleet review inbox, in the browser

Approve a proposal from your phone over ntfy

Watch and govern the whole fleet from your browser, and approve from your phone.

See the extension →

Explore the product →

Built on open source

The free engine the product is built on. It runs alongside the AI coding tools you already use, not instead of them. Start local, connect when you grow.

All open source →

empirica

MIT

The framework that gives a single AI its track record and calibration, on your own machine. Think before acting, checked against real evidence.

What it does GitHub

ecodex

Apache-2.0 alpha

The same discipline for the OpenAI Codex CLI and any OpenAI-compatible model, so the AI will not act past what it actually knows.