Flagship Product

Agent Protocol Layer

A shared symbolic protocol between AI agents that makes every step of A2A commerce — reasoning, negotiation, execution — formally verifiable. Not a communication channel. A constraint layer that eliminates hallucination.

See DemoArchitecture →
The Problem

Agents can talk. But can you trust them?

Two LLM agents can negotiate in natural language out of the box. But every inference is a black box — unverifiable pricing logic, untraceable inventory claims, hallucinated terms. In low-stakes chat this is fine. In real commerce with real money, it's unacceptable.

Buyer Agent Seller Agent ? ? ?
The Solution

Shared symbolic protocol as formal constraint

The Jachin Protocol Layer is not a communication middleware. It's a shared logical grammar — a formal symbolic system that both agents reason through. Every claim is type-checked against the protocol. Every inference follows declared rules. Every conclusion carries a proof chain.

Buyer Agent Seller Agent PROTOCOL
Scenario — EE Coffee

One buyer, three suppliers. Full audit trail.

Watch how the EE Coffee Agent uses the protocol layer to simultaneously negotiate with three supplier agents while performing internal reasoning on seasonality, inventory burn rate, and cash flow — then selects the optimal deal with a fully traceable decision chain.

Step 01

Inventory Trigger

Arabica stock drops below the reorder threshold. The EE Agent detects the shortfall through continuous inventory monitoring and initiates the procurement workflow.

Protocol: inventory_level(arabica) < reorder_threshold → trigger(procurement_workflow)
Buyer Agent
Step 02

Internal Reasoning

Before contacting any supplier, the EE Agent reasons over its own context: current season and demand patterns, inventory burn rate, and cash flow constraints. The protocol layer enforces formal deduction — the resulting price range and quantity are derived conclusions, not guesses.

Protocol: season(rainy) ∧ demand_forecast(+15%) ∧ cash_reserve(tight) → target_price_range($12.40–$13.80/kg) ∧ order_qty(200kg)
Buyer Agent
Step 03

Parallel Negotiation

The EE Agent opens simultaneous negotiations with three supplier agents. Each conversation runs through the shared protocol layer — semantically isolated but formally consistent. No cross-contamination of terms. No misattributed quotes.

Protocol: ∀ supplier ∈ {A, B, C}: negotiate(supplier, target_price, qty) → session_isolated ∧ terms_typed ∧ state_tracked
Buyer Agent Seller A Seller B Seller C
Step 04

Convergent Decision

All three offers return. The EE Agent doesn't just compare prices — it performs multi-factor logical deduction across delivery terms, quality guarantees, payment flexibility, and its own constraints. The decision is a derived conclusion, not a ranking.

Protocol: evaluate(A: $13.20, 5d) ∧ evaluate(B: $12.60, 7d, flex_pay) ∧ evaluate(C: $12.90, 3d) ∧ cash(tight) → select(B) ∵ flex_pay ∧ price_in_range
Seller A Seller B ✓ Seller C Buyer Agent
Step 05

Execution & Proof

Order confirmed. Payment triggered. The entire decision chain — from inventory shortfall to supplier selection to final terms — is recorded as a verifiable proof chain. Any step can be audited: why this supplier, why this price, why this quantity.

Protocol: execute_order(B, 200kg, $12.60) → proof_chain[trigger → reasoning → negotiation × 3 → selection → execution] ∎
What the Protocol Layer Does

Four constraints that eliminate hallucination

Each capability is a formal guarantee, not a probabilistic improvement. The protocol layer transforms unverifiable agent dialogue into auditable logical transactions.

Semantic Alignment

Both agents share the same symbolic definitions. "Arabica," "delivery window," "unit price" mean precisely the same thing on both sides. No ambiguity, no drift.

Inference Rules

Every reasoning step follows declared logical rules. "If cash is tight AND supplier offers flexible payment → increase preference weight" is an explicit rule, not an emergent pattern.

State Tracking

The protocol layer tracks the full state machine of each negotiation — which terms are proposed, countered, accepted. Three parallel sessions, zero state confusion.

Proof Chain

The complete decision path — from trigger to execution — is recorded as a formal proof. Auditable at every node. "Why Supplier B?" has a logical answer, not a statistical one.

Why Jachin

Not middleware. Not smart contracts. Ontology.

Other approaches to agent-to-agent coordination — API middleware, smart contracts, prompt engineering — all share the same limit: they constrain behavior without understanding structure. Jachin's protocol layer works because underneath it sits a formal model of the world. The rules aren't arbitrary constraints. They derive from how things actually relate. That's why the protocol can evolve: as the ontological layer deepens, agents don't just follow better rules — they emerge their own.

API Middleware

Routes messages. No semantic understanding. Can't verify reasoning. Can't evolve.

Smart Contracts

Enforces pre-coded terms. No reasoning. Can't handle ambiguity. Can't learn.

Jachin Protocol

Reasons on world structure. Verifies every step. Evolves as ontology deepens. Rules emerge — not imposed.

Evolution Path

From symbolic rules to full ontology

The protocol layer is designed to deepen over time. Begin with explicit symbolic rules that handle core procurement logic. As the ontological layer matures, agents gain richer contextual reasoning — without breaking existing workflows.

Phase 1 — Symbolic

Explicit inference rules handle core procurement logic. Deterministic, auditable, deployable now.

  • Type-checked semantic alignment
  • Rule-based price/quantity reasoning
  • State machine negotiation tracking
  • Full proof chain output
.

Phase 2 — Full Ontology

Full ontological layer enables richer contextual reasoning. Agents understand not just rules, but the structure of the domain itself.

  • Substance-accident distinction in entities
  • Causal reasoning (four causes)
  • Seasonal/contextual inference from structure
  • Cross-domain functor mapping
.
Next Step

See the protocol in action