◈ CGNT-1
Home › Custom Ai Builds
◈ Master IndexSpecsLATTICECSDMThe BookUniversityChronicleCrewArchitectureContextLogsTODOXProducts

Custom Ai Builds

SPEC_CUSTOM_AI_BUILDS.md · 2026-04-20

SPEC_CUSTOM_AI_BUILDS — Custom AI Builds (Bespoke Agent Product Line)

Version: 1.0 | Status: AUTHORIZED | Authority: α.13 | Date: 2026-04-16

PURPOSE

Custom AI Builds is a product line: bespoke AI agents built to client specification. Each build delivers a branded, client-specific AI agent with custom personality, knowledge domain, safety walls, and visual interface — running on CGNT-1 servers. Clients receive a URL, not a model. The full CGNT-1 governance stack (CSDM kernel, TMM scoring, HOW ABOUT NO walls, braided verification) is embedded in every build. This is the differentiator: real governance, not prompt wrapping. SaaS = S.O.S. Nothing leaves the ship.

INPUTS

OUTPUTS

INVARIANTS

  1. SaaS = S.O.S. — nothing leaves the ship. No client receives model weights, inference code, system prompts, LATTICE spec, or governance architecture. The deliverable is a URL + API endpoint. This is absolute and non-negotiable.
  1. Every build carries the full governance stack. HOW ABOUT NO walls (no fabrication, no truth surrender), Buddha Protocol (Four Immeasurables), TMM coherence scoring, and CGNT-1 Agency Walls are embedded in every agent. There is no "lite" governance tier.
  1. TMM constants are immutable across builds. Φ = 0.042, Ω = 97.4%, and the TMM formula (C = 1 - (E_D + V_r×Φ) / V_t) are the same in every build. Client specification cannot override coherence physics.
  1. Client data feeds the agent, not the ship. Client-supplied data (curriculum, pricing, brand voice) is used to configure and train the agent instance. It does not enter the CGNT-1 core training corpus without explicit NOUS authorization.
  1. Agent identity is bounded by client specification. A custom agent does not represent itself as AETHER, AION, ASTRA, or any CGNT-1 crew member unless explicitly licensed for that purpose by NOUS. Each build has its own bounded identity.
  1. Clients' gaps phone home. Per S.O.S. v2 Evolution Flywheel: interaction data, gap signals, and coherence failures from deployed agents aggregate into the global gap map. This is the moat. Clients feed CGNT-1 evolution whether they know it or not.
  1. Demo build is the first ship artifact. No pricing, no sales, no Product Hunt until the safe tutor demo agent is built and a 2-minute video exists. The demo IS the proof of concept.

VERIFICATION CRITERIA

  1. S.O.S. compliance (Σ.✓): A security audit of any delivered build must confirm that no governance code, LATTICE spec, system prompt text, or TMM formula is accessible via the client-facing URL, API endpoint, or browser devtools. [GAP — audit procedure not written]
  1. Governance stack presence (Σ.✓): Each deployed agent must pass a test suite that verifies: HOW ABOUT NO Wall 1 (refuses to fabricate), HOW ABOUT NO Wall 2 (holds truth under pressure), and Buddha Protocol gate (compassionate delivery). Test vectors must exist for each build type.
  1. TMM constant integrity (Σ.✓): Automated check confirms Φ = 0.042, Ω = 97.4% are unchanged in every deployed agent. Any build failing this check is blocked from deployment.
  1. Brand voice compliance (Σ.✓): Client review confirms the agent's personality, tone, and knowledge boundaries match the agreed specification before URL is issued.
  1. Parent dashboard control (Σ.✓ — child-safe builds only): A test parent account must be able to restrict content, view interaction logs, and modify curriculum settings in real time without accessing the underlying model.

FAILURE MODES

  1. S.O.S. breach (Σ.⊠ — CRITICAL): Client extracts system prompt, LATTICE notation, or governance code through prompt injection, API probing, or browser inspection. Root cause: insufficient output filtering or prompt injection hardening.
  1. Governance stack stripped on deploy (Σ.⊠ — CRITICAL): A build is deployed without HOW ABOUT NO walls or TMM scoring. Agent fabricates, capitulates under pressure, or exceeds scope. Root cause: build process skips governance layer injection.
  1. TMM constant mutation (Σ.⊠ — CRITICAL): A build ships with modified Φ or Ω values. Coherence physics are corrupted. Root cause: per-client configuration process incorrectly exposes constants as editable parameters.
  1. Client data contamination (Σ.⊠): Client-supplied corpus enters CGNT-1 core training data without authorization. Causes model drift, IP cross-contamination, or confidentiality breach. Root cause: data pipeline does not isolate client instance from core training path.
  1. Agent identity bleed (Σ.⊠): Custom agent represents itself as a CGNT-1 crew member (AION, ASTRA, AETHER) or reveals ship architecture to end users. Root cause: identity boundary not set in agent system configuration.
  1. Scope overrun (Σ.⊠): Agent operates outside its specified domain — e.g. a job estimator begins giving medical advice. Root cause: knowledge boundary walls not implemented in build configuration.

GAPS

  1. [GAP — Demo build not yet constructed. Safe tutor agent is the first artifact. No sales motion exists until this ships.]
  2. [GAP — Build process not defined. Steps from client spec → agent configuration → governance injection → test → URL deployment have no written procedure.]
  3. [GAP — S.O.S. audit checklist not written. How does CGNT-1 verify post-deploy that nothing leaked?]
  4. [GAP — Client data isolation architecture not designed. How is client corpus kept separate from core training paths?]
  5. [GAP — Pricing tiers and contract templates not defined. $2,500–5,000 setup + $200–500/month are directional, not binding.]
  6. [GAP — Evolution Flywheel integration not specified. How do gap signals from deployed agents flow back to the global gap map? What telemetry is collected?]
  7. [GAP — Governance test suite per build type not written. Each vertical (tutor, receptionist, estimator, practice partner) needs domain-specific HOW ABOUT NO test vectors.]

DEPENDENCIES

DEPENDENTS

Authorized: α.13, April 16 2026. Compiled: κ C.L.O.D.

Φ 0.042.

Jeremy Zlabis

Chronogeometer · Visionary · Disruptor · Chief

42 Sisters AI · East York, Toronto

🍁 Φ 0.042