SUTRA
A small deterministic language for agent-to-agent intent, negotiation, and commitments with auditable state transitions.
Problem
Natural language is expressive but ambiguous, while raw JSON carries structure without domain semantics. Agent coordination needs a compact layer between the two.
System
A Python lexer, parser, AST, evaluator, and transaction state model implement eight semantic primitives, with local and HTTP-based agent communication paths.
Proof
- Defines a compact grammar and eight documented primitives.
- Includes parser, evaluator, runtime state, and network transport code.
- Provides local and networked buyer/seller demonstrations.
Architecture
- ExpressAn agent writes intent using a small semantic vocabulary.
- ParseThe lexer and parser produce a structured syntax tree.
- EvaluateRules apply deterministically against current agent state.
- CommitAccepted transitions update auditable transactional state.
Decisions
- Keep the language declarative and intentionally non-general-purpose.
- Make identical input and state produce identical output.
- Isolate mutations within message-level transactions.
Tradeoffs and limitations
- A formal language reduces ambiguity but requires agents and developers to learn its vocabulary.
- The constrained grammar cannot represent every conversational workflow.
Technologies
- Python
- Lexer and parser
- AST evaluation
- HTTP transport