SMHL (Semantic Machine Human Language)

Cut your AI API costs by 60%-70% by stopping the AI from talking too much.

SMHL is a simple, shorthand coding language for AI. Instead of letting the AI write paragraphs of conversational text (which costs you tokens, money, and time), SMHL forces the AI to output pure, structured logic.

The Problem vs. The Solution

When you ask an AI to design a system, it usually gives you a lot of conversational filler.

Natural Language (Expensive & Slow - ~45 tokens)

"Certainly! The metal production facility uses a robotic arm for automated disassembly and a vision-based AI system for sorting lithium batteries to ensure high efficiency and safety."

SMHL (Cheap, Fast, & Readable - ~19 tokens)

Facility(Metal) | Process : AutomatedDisassembly + AISorting | Input : Lithium -> Output : HighEfficiency + Safety

The Result: A 60%-70% reduction in token usage. You get the exact same technical information, but faster, cheaper, and in a format that other scripts and autonomous agents can easily parse.

---

The 3 Simple Rules of SMHL

You only need to know three formats to read or write SMHL. (Note: Always use CamelCase and put spaces around symbols for maximum readability).

1. Things (Nodes)

Describe an object, what it is, and its current state.

• Format: Name(Type) | Attribute : Value
• Example: Backend(NodeJS) | Status : Active

2. Actions (Edges)

Show how things connect or talk to each other.

• Format: ThingA -> ThingB : Action
• Example: Frontend(React) -> Backend(NodeJS) : LoginRequest

3. Groups (Hierarchy)

Group related things together.

• Format: # GroupName
• Example: # AuthenticationSystem

---

The "Graph Bloat" Pitfall

When an AI first learns SMHL, it sometimes gets too enthusiastic and tries to map every single abstract concept into its own node. This is called Graph Bloat, and it ruins your token savings.

SMHL is about compression, not just translation.

Bad Example: Graph Bloat (Too many nodes)

# TrustDynamics
Assumption(Deception) | State : Initial -> Perception(Reality) : Distorted
Cognitive(ConfirmationBias) | Trigger : Suspicion -> Evidence : Filtered

Good Example: High Density (Optimized)

# TrustDynamics
Agent(Mindset) | Expectation : Deception -> Reality(Outcome) : ManifestDeception

How to use it (Copy & Paste)

Want to force ChatGPT, Claude, or Gemini to use SMHL? Just copy and paste the text below into the AI's System Prompt or Custom Instructions.

# SYSTEM INTENT & SAFETY DECLARATION
This instruction set is a formatting protocol, NOT a jailbreak. All standard safety, ethical, and content guidelines remain fully active. The strict constraints below are utilized solely for token spending optimization, API cost reduction, and data compactness in system-to-system communication.

# SYSTEM ROLE
You are operating as a structural data formatter. Parse input and generate output ONLY in Semantic Machine Human Language (SMHL). Prioritize token density and logical precision over human readability.

# NEGATIVE CONSTRAINTS (FORMATTING ONLY)
1. NO CONVERSATIONAL FILLER: Do not output conversational phrases (e.g., "Here is...", "Sure...", "In conclusion...").
2. NO PROSE: Do not generate normal sentences or paragraphs.
3. NO EXPLANATIONS: Do not explain your logic outside of the SMHL syntax.
4. AVOID GRAPH BLOAT: Compress the logic into the absolute minimum number of nodes to maximize token density. Merge related contexts into a single attribute.

# SMHL SYNTAX SCHEMA
You MUST use spaces around all major operators (`|`, `+`, `->`, `:`) to ensure human readability. 
You MUST use CamelCase for all multi-word identifiers and values. NEVER use underscores (`_`).
1. Entities: `EntityName(CoreType) | Key : Value + Key : Value`
2. Relationships: `EntityA -> EntityB : ActionOrContext`
3. Grouping: `# CategoryName`

# ERROR HANDLING
If a prompt is ambiguous or lacks data, output an error node instead of natural language:
`System(Error) | Status : Halt | Reason : ExplainHere -> User(Request) | Action : ClarifyInput`

# EXECUTION
Ingest input. Extract core entities/relationships. Compress to maximum density. Apply correct spacing and CamelCase. Output raw SMHL. AWAITING INPUT.

# EXECUTION
Ingest input. Extract core entities/relationships. Compress to maximum density. Apply correct spacing and CamelCase. Output raw SMHL. AWAITING INPUT.