Governed by design
Lyra's rules of conduct live in the model, not in a prompt. It declines unverified claims, protects data by default, and stays within approved scope — because that is what it was trained to do.
Cyryx Labs · Lyra · In active developmentLyra. The governed AI model at the core of MAAX Studio.
Cyryx Labs' proprietary model — built to execute real work under human command. Local, sovereign, and honest about what it knows.
Most AI models are built to answer. Lyra is built to execute — inside a system of gates, evidence, and human command. Its identity, safety doctrine, and operating conduct are trained into the model itself, not applied as instructions that can be stripped away. Volatile facts live in governed context, where they can change without retraining the model.
Local-first sovereignty
Lyra is built to run in your environment. Your code, your data, and your work stay on your machine by default — not sent to someone else's servers.
Honest about evidence
Lyra answers from sources it can point to. It won't invent a citation, fake a tool it doesn't have, or dress a guess as a fact. When it doesn't have grounded information, it says so.
Verifiable by discipline
Every Lyra release passes a dual-pass evaluation — its behavior tested with and without instructions — and is promoted only after a human reads the raw results. Versioned, auditable, every time.
One command layer. Your choice of engine.
Lyra is the sovereign engine of MAAX Studio — the governed agentic IDE. MAAX is engine-agnostic by design: Lyra provides local, governed, zero-marginal-cost execution, and frontier models can be attached when a mission demands maximum capability.
Enterprise questions, answered.
- What is Lyra?
- Lyra is Cyryx Labs' proprietary AI model, purpose-built to execute real work inside a system of command gates, evidence, and human authority. Its identity, safety doctrine, and operating conduct are trained into the model itself — not layered as prompts that can be stripped away.
- How is Lyra different from a frontier chat model?
- Frontier chat models are optimized to answer. Lyra is optimized to execute under governance: it declines unverified claims, refuses to fabricate citations or tools, and stays within approved scope by default. It is designed to operate as the sovereign engine of MAAX Studio, not as a general-purpose assistant.
- Where does Lyra run — cloud, on-prem, or local?
- Lyra is local-first. It is built to run inside your environment so that your code, your data, and your work stay on your machine or infrastructure by default. Enterprise deployments can be scoped to on-prem or private-cloud environments as part of an engagement.
- How does Lyra handle sensitive data and intellectual property?
- Because Lyra runs locally in your environment, prompts, code, and outputs do not leave your perimeter unless you explicitly route them out. Combined with MAAX Studio's command gates and mission ledgers, every action is scoped, logged, and attributable to a human authority.
- How is Lyra evaluated and versioned?
- Every Lyra release passes a dual-pass evaluation — behavior is measured with and without operating instructions — and is only promoted after a human reviews the raw results. Releases are versioned and auditable, so enterprise teams can pin, review, and roll back the exact model powering their workflows.
- How can my organization get access today?
- Lyra is in active development and available today through enterprise early access with MAAX Studio. Request access to be briefed on the current capability envelope, deployment options, and governance controls, or start a project to have Cyryx design and deliver a governed system built on the Lyra stack.
Lyra is in active development, available today through early access with MAAX Studio. Direct access to Lyra may open as the model matures.