Not prompts. Not fine-tuning. Governance contracts that make AI auditable and accountable.
A CogniMap is not a prompt. It's a governance contract — a structured specification that defines how an AI must behave, what evidence it can reference, and when it must declare uncertainty.
When you interact with a CogniMap-powered AI, you're not talking to a chatbot that's been told to "act like an expert." You're talking to an AI operating under a binding contract that enforces verification requirements, evidence citation, and behavioral boundaries.
The key difference: Traditional AI can hallucinate freely. A CogniMap-governed AI must reference its evidence registry and explicitly declare when claims are unverified.
When you select a CogniMap, the system loads that governance contract server-side and injects it into the AI's core instructions. This means:
Maps are never exposed to the client — your governance logic stays private.
New maps can be added without redeploying the system.
The AI references the map's evidence registry during responses.
Researcher Mode reveals execution traces for full transparency.
This is fundamentally different from pasting a prompt into ChatGPT. The contract governs the AI at an architectural level.
| Aspect | System Prompt / Custom GPT | CogniMap |
|---|---|---|
| Nature | Personality instruction | Governance contract |
| Verification | None — AI can claim anything | Every claim has explicit status |
| Evidence | None — no source tracking | Dual registry linking claims to evidence |
| Auditability | Black box | Researcher Mode shows traces |
| Updates | Requires redeployment | Hot-swappable JSON files |
| IP Protection | Client-visible prompts | Server-side only |
Every production CogniMap maintains two registries that work together to ensure accountability:
What the system believes to be true, with support levels:
What supports each claim:
This architecture means the AI can't just make things up. It must trace claims back to evidence and declare when that evidence is missing.
The CogniMap format is substrate-agnostic — the same governance contract produces identical expert behavior across different AI architectures. This has been validated across frontier models:
Claude 4 family
GPT-4 family
Gemini family
Grok family
V3 & R1 family
Llama family
Large & Small
Liquid AI, Qwen
The marketplace currently runs on Google Gemini Flash, but the underlying CogniMap format works identically across all architectures. This portability makes CogniMaps a new category — not prompts, not fine-tuning, but governance contracts that transfer across AI substrates.
The marketplace uses a credit system to manage API costs:
| Package | Price | Credits | Per Message |
|---|---|---|---|
| Starter | $5 | 50 credits | $0.10 |
| Standard | $10 | 120 credits | $0.08 |
| Pro | $25 | 350 credits | $0.07 |
Each chat message costs 1 credit. Payments processed securely via Stripe.
Need AI responses they can trust and trace — not black-box outputs.
Require auditable reasoning with explicit evidence linkage.
Every AI claim must be verifiable for regulatory requirements.
Tired of AI that confidently makes things up.
Experience AI with accountability. Select a domain expert and start chatting.
"The AI can't just hallucinate. It must reference the evidence registry and declare when claims are unverified."— Higgs AI