Platform

NEO AI owns the Brain. Models are interchangeable workers, addressed through one contract, and replaceable without a rewrite [A]. The pages below describe the parts of that system that are specified in the public corpus, each carrying the status the Capability Register gives it — and nothing carries a status the register does not.

Architecture v1.0 is frozen at 17 components, following Amendment 001, which reinstated Secure Execution after a deletion argument was tested against the receiving files and found false [A]. The count is a fact about the specification, not a claim about a running system.

What is published here

• Data Plane — where an answer's inputs come from, under which licence, and what the record of that is. Four parts, one capability group [D] (data-plane, S4).
• The efficiency programme — the arithmetic that would have to hold before any magnitude claim could be made, and the rung the Program is licensed to stand on today, which is none.
• NEO System Intelligence — The architectural layer above models, agents, tools, memory, knowledge, evidence and evaluation. [D] (brain, S4)
• LILI and the specialists — The primary intelligence identity. One interface, many specialists. [D] (lili, S4)
• Evidence, memory and provenance — Where an answer came from. [D] (brain.evidence-layer, S4) [D] (brain.memory, S4)
• Human authority — Who decided, and were they entitled to. [D] (brain.governance, S4) [D] (mission.envelope, S4)
• Security-first architecture — Security is a filter, not a weight.
• Deployment boundaries — One system, several boundaries. [D] (deployment.cloud, S4) [D] (deployment.vpc, S5) [D] (deployment.onprem, S5)
• Mission Control — Supervision as a product: every rendered element resolves to a persisted event about its own subject [D] (mission-control, S4)
• Observe Build — The build view of Mission Control: the same stream, the same projection, the same rules [D] (mission-control, S4)

What is deliberately not published here

No benchmark. No comparison against another system. No latency, accuracy or uptime figure for any source, because those belong to Source Intelligence, which is a view over an Execution Store that has not yet run outside the Program's own harness [D] (data-plane.source-intelligence, S5).