Crate frame_isa 

Frame ISA - SAM Instruction Set Architecture

Zero-dependency crate defining the 6-byte opcode format for SAM AI systems.

Overview

Frame ISA defines a compact instruction format for AI output:

[ACT:2 bytes][SUBJ:2 bytes][MOD:2 bytes] = 6 bytes total (big-endian)

• ACT (Action): What operation to perform (GREET, RESPOND, CALCULATE, etc.)
• SUBJ (Subject): The topic or entity (TIME, USER, WEATHER, RAG reference, etc.)
• MOD (Modifier): Style flags (voice, tone, warmth, format, urgency, etc.)

Usage

use frame_isa::{Action, Subject, Modifier, Instruction};

// Create an instruction
let instr = Instruction::new(
    Action::RESPOND,
    Subject::TIME,
    Modifier::default(),
);

// Serialize to bytes
let bytes = instr.to_bytes();
assert_eq!(bytes.len(), 6);

// Parse from bytes
let parsed = Instruction::parse_one(&bytes).unwrap();
assert_eq!(instr, parsed);

// Use the builder
use frame_isa::{InstructionBuilder, Voice, Tone};

let instr = InstructionBuilder::new(Action::GREET)
    .subject(Subject::USER)
    .voice(Voice::Casual)
    .tone(Tone::Positive)
    .build();

Action Categories

Range	Category	Examples
0x00xx	System	NOP, HALT, ERROR, STATUS
0x01xx	Response	GREET, CONFIRM, DENY, EXPLAIN, RESPOND
0x02xx	Query	ASK, REQUEST, SEARCH, RETRIEVE
0x03xx	Knowledge	DEFINE, DESCRIBE, COMPARE, SUMMARIZE
0x04xx	Skill	CALCULATE, SET_TIMER, KNOWLEDGE_SEARCH
0x05xx	Emotion	EMPATHY, CONCERN, ENCOURAGEMENT
0x06xx	Template	TEMPLATE_LOAD, TEMPLATE_FILL
0x07xx	Chain	CHAIN, FORK, MERGE

Subject Categories

Range	Category	Examples
0x00xx	System	NULL, SELF, USER, CONTEXT
0x01xx	Common	WEATHER, TIME, DATE, SCHEDULE
0x02xx	Math/Science	NUMBER, EQUATION, PHYSICS
0x03xx	Technology	COMPUTER, SOFTWARE, AI, API
0x04xx	Knowledge	DOCUMENTATION, CONCEPT
0x05xx	Emotions	FEELINGS, STRESS, ANXIETY
0x06xx	TRM Refs	References to other TRM models
0xE0xx	RAG Refs	Dynamic document lookups

Modifier Bit Layout

Bit:  15  14  13  12  11  10   9   8   7   6   5   4   3   2   1   0
      [--VOICE--] [--TONE--] [-WARM-] [--FORMAT--] [ACCURACY] [URGENCY]

Integration with TRM Models

TinyRecursiveModels (TRMs) output predictions that map to these opcodes. The factored prediction approach uses 3 heads:

• ACT head → Action code
• SUBJ head → Subject code
• MOD head → Modifier flags

This allows small models (~148K params) to achieve 99%+ accuracy on opcode prediction tasks.

Re-exports

pub use action::Action;

pub use instruction::Instruction;

pub use instruction::InstructionBuilder;

pub use instruction::InstructionError;

pub use instruction::INSTRUCTION_SIZE;

pub use modifier::Accuracy;

pub use modifier::Format;

pub use modifier::Modifier;

pub use modifier::Tone;

pub use modifier::Urgency;

pub use modifier::Voice;

pub use modifier::Warmth;

pub use subject::Subject;

Modules

action

Action codes for the SAM ISA

instruction

Instruction representation for the SAM ISA

modifier

Modifier flags for the SAM ISA

prelude

Convenience prelude for common imports

subject

Subject codes for the SAM ISA

Constants

ISA_VERSION

Current ISA version