FAQ
Questions, answered.
General
What is Matrix?
Matrix is an agent cognition platform. It provides the infrastructure layer between natural language and system execution — routing every action through a typed Intent IR before anything happens in the world.
What is the difference between Neo and MCL?
Neo is the conversational rail for reversible everyday work. MCL is the rigorous rail for irreversible and high-stakes operations. Neo automatically delegates to MCL when an action is irreversible.
Does Matrix require a specific LLM?
No. Matrix is model-agnostic. The Intent IR layer works with any language model capable of structured output. We provide optimized integrations for the major frontier models.
Security & Trust
How does Matrix prevent agents from taking unauthorized actions?
Every action must first be represented as a typed Intent IR. The IR is validated against a permission set before the executor walks it. Irreversible actions require MCL sign-off, which enforces a closed verb vocabulary and cannot be bypassed by the model.
Are agent credentials stored securely?
Yes. The UWAC layer handles OAuth vaulting on a per-user, per-tool basis. Credentials are never passed through the model context window.
Technical
What is the Intent IR?
The Intent IR (Intermediate Representation) is a protocol that maps natural language to a structured object with a typed verb, typed objects, and a declared reversibility class. It acts as a compiler layer between language and execution.
Can I inspect what my agent is about to do before it happens?
Yes. The Intent IR is human-readable and API-accessible. You can hook into the pre-execution checkpoint to review, modify, or cancel any planned action.