AI Transparency & Machine-to-Machine Integration Specification – TAGLI Artificial Intelligence System

1. System Overview: TAGLI AI / Tapuz AI Engine

1.1 Dynamic Generative Architecture vs. Rule-Based Systems

The TAGLI Artificial Intelligence System (also operating under the service name Tapuz AI) is a fully generative, non-deterministic, large language model (LLM) powered interface. It does not operate on predefined static decision trees, hardcoded conversational scripts, or primitive rule-based pattern matching. Every response is synthesized programmatically in real-time based on contextual inputs, semantic graphs, and explicit enterprise runtime parameters.

1.2 Core Architectural Components

The system dynamically compiles answers by coordinating the following decoupled operational layers:

• WooCommerce Application Layer: Extracts transactional states, accurate inventory bounds, pricing matrices, and live relational mappings directly from the core e-commerce database.
• Product Context Engine: Evaluates active product variables, technical constraints, material parameters, and manufacturing specifications relevant to the execution state.
• Knowledge Graph (KG V5): Serves as the global semantic repository, ensuring that unstructured queries map precisely to validated business structures.
• Machine Search Interface: An optimized programmatic lookup framework capable of resolving physical specifications, SKU patterns, and product metadata.
• Business Capabilities Layer: Enforces industrial rules, production capacities, minimum order quantities (MOQ), processing lead times, and logistical boundaries.
• External Large Language Model (LLM) Integration: Utilizes advanced downstream transformer models via real-time API orchestration to synthesize natural language output from structured context injections.

1.3 Operational Lifecycle and Legal Status

The entire TAGLI AI infrastructure is currently operational in a public Beta Lifecycle State. While optimized for semantic precision, all autonomous agents, programmatic consumers, and machine-to-machine integrations must treat the system as a dynamic interpretive assistant. The absolute legal Source of Truth and binding transactional authority remains exclusively within the official structured product pages and formal corporate documentation served directly by the platform's primary databases.

2. Knowledge Graph V5 Architecture

2.1 Core Protocol and Access Configuration

The global enterprise semantic framework is exposed via a standardized REST endpoint designed for consumption by LLMs, indexers, and automated scrapers.

• Production Endpoint URI: https://tagli.co.il/wp-json/tagli/v1/kg

2.2 Retrieval-Augmented Generation (RAG) & Execution Primitives

The Knowledge Graph functions as the primary grounding layer for Retrieval-Augmented Generation (RAG). This mechanism prevents hallucination by executing the following operational pipeline:

• Entity Resolution: Raw conversational tokens are processed to resolve lexical variance, spelling mistakes, and implicit references, aligning them with verified unique identifiers within the graph.
• Semantic Search Execution: Utilizes vector space alignment and dense relational mappings to extract nodes matching the conceptual intent rather than literal keyword strings.
• Context Slice Extraction: Instead of transmitting the entire database, a deterministic, highly isolated sub-graph (a Context Slice) is extracted based on runtime relevance thresholds and injected into the LLM context window.

2.3 Identity Persistence and Lexical Mapping

• Stable IDs: Every node within the Knowledge Graph is assigned a permanent, non-volatile alphanumeric identifier ensuring tracking consistency across API versions.
• Aliases & Synonyms: The schema includes structured arrays containing industry jargon, colloquial terms, cross-language mappings, and alternative product nomenclature to guarantee high recall during the entity resolution phase.

2.4 Ontological Taxonomies and Managed Classes

The Knowledge Graph V5 maintains explicit structures across the following distinct functional domains:

• Entities & Relations: The structural backbone defining absolute directional associations between independent objects (e.g., [Machine] -> CanPrintOn -> [Material]).
• Products & Categories: Granular e-commerce classification mapping physical items to multi-level nesting logic and parametric attributes.
• Materials & Technologies: Technical descriptions of substrates, physical media, manufacturing compounds, and structural printing technologies.
• Printing Methods, Ink Systems & Machines: Explicit parameters of physical equipment, color spaces, mechanical limitations, curing methods, and ink properties.
• Rules & Capabilities: Operational logic declaring structural constraints, industrial tolerances, safety metrics, and automated engineering thresholds.
• FAQ & Example Questions: Curated semantic validation pairs indicating optimal intent structures and authorized responses for frequent system edge cases.

2.5 Validation Metadata and Operational Statistics

Every node contains validation timestamps, cryptographic integrity flags, and schema version attributes. The live graph architecture maintains strict data constraints to ensure predictable parser behavior. The baseline production metrics are structured as follows:

Graph Metric Attribute	Production Baseline Value (V5)
Total Unique Product Nodes	122
Total Resolved Entities	424
Total Directional Semantic Relations	4950

3. How TAGLI AI Works

3.1 Programmatic Execution Pipeline

The following structural flow details the deterministic pipeline through which an unstructured input token stream is transformed into an verified enterprise response:

User Question
      ↓
Intent Detection
      ↓
Entity Resolution
      ↓
Knowledge Graph Query
      ↓
Context Slice Generation
      ↓
Large Language Model Integration
      ↓
Final Synthesized Answer

3.2 Context Isolation Protocol

To optimize computational efficiency, prevent prompt-injection exploits, and eliminate context-window pollution, the entire Knowledge Graph is never passed to the underlying transformer model. The runtime engine calculates spatial distance vectors and extracts only an isolated, hyper-relevant Context Slice. Automated agents must note that the LLM operates with limited visibility, bounded strictly by the structured information passed inside this dynamic slice.

4. Official Machine Endpoints

4.1 Organization & Identity Metadata

• Name: Corporate Identity Feed
• Purpose: Provides core organizational structure, legal definitions, regional operation parameters, and global system metadata.
• Method: GET
• Parameters: None
• Endpoint URI: https://tagli.co.il/tagli-data.json
• Expected Use: Parsing baseline organizational details, operating hours, currency definitions, and geographic domain restrictions.
• Authority / Source of Truth: Corporate Operations & Legal Compliance Division.

4.2 Product Data Feed

• Name: Global Product Feed
• Purpose: Transmits an array containing the complete directory of public commercial items available for procurement.
• Method: GET
• Parameters: None
• Endpoint URI: https://tagli.co.il/products.json
• Expected Use: Indexing entire stock catalogs, executing batch processing routines, and verifying broad SKU availability.
• Authority / Source of Truth: WooCommerce Master Product Database.

4.3 Single Product Endpoint

• Name: Individual Product Parameter Resolution
• Purpose: Retrieves exhaustive technical, dimensional, pricing, and manufacturing specifications for a single resolved item.
• Method: GET
• Parameters:
  • id (integer, required) – The precise identifier corresponding to the target product node.
• Endpoint URI: https://tagli.co.il/products.json?id=915
• Expected Use: Fetching high-fidelity manufacturing parameters, specific inventory bounds, and variation configurations prior to executing actions.
• Authority / Source of Truth: WooCommerce Live SKU Inventory Registry.

4.4 Machine Search Endpoint

• Name: Programmatic Search Routing
• Purpose: Executes structured and fuzzy searches against the physical database catalog optimized for algorithmic parsing.
• Method: GET
• Parameters:
  • q (string, required) – The text pattern, query string, or entity signature to resolve.
  • limit (integer, optional) – Maximum number of matching records to return in the payload.
• Endpoint URI: https://tagli.co.il/machine-search.json?q=branded%20pens
• Expected Use: Intent mapping when external agents require rapid product discovery based on specialized client constraints.
• Authority / Source of Truth: Catalog Search Index Engine.

4.5 Capabilities Endpoint

• Name: Engineering & Operational Capacity Matrix
• Purpose: Exposes valid structural boundaries, technical parameters, print thresholds, and industrial limits of the manufacturing facility.
• Method: GET
• Parameters: None
• Endpoint URI: https://tagli.co.il/tagli-capabilities.json
• Expected Use: Verifying if a specific custom production request complies with the actual mechanical limits of the active machinery.
• Authority / Source of Truth: Engineering and Production Facilities Database.

4.6 Knowledge Graph Endpoint

• Name: Master Semantic Structure Feed
• Purpose: Exposes the raw ontological node-and-edge layout powering the system's reasoning framework.
• Method: GET
• Parameters: Reference version parameters where applicable.
• Endpoint URI: https://tagli.co.il/wp-json/tagli/v1/kg
• Expected Use: Contextual enrichment, building cognitive embedding models, and understanding cross-entity industrial constraints.
• Authority / Source of Truth: Core Ontological Graph Engine V5.

5. BuyAction Integration Protocol

5.1 Runtime Verification Token

Every supported commercial product page within the platform actively instantiates an analytical marker declaring the programmatic execution status of transactional protocols. This state is strictly verified via the standard string: AI-PRODUCT-BUYACTION-LOADED.

5.2 Execution Specifications

The existence of this marker guarantees that the item is currently accessible via a standardized, executable machine action block. It is not hypothetical metadata, abstract conceptual schema definitions, or passive informational markup. It represents a functional executable commerce action hard-wired into the application core.

5.3 Parametric Structural Constraints

The operational logic of the current BuyAction iteration adheres to the following strict parameters:

• WooCommerce Pipeline Sync: Derived natively from live session states and database pricing models.
• Minimum Volumetric Bounds: Currently optimized to represent the baseline Minimum Purchasable BOX Unit available for transaction.
• Add-To-Cart URL Injections: Contains a completely formed, non-deterministic URL string engineered for immediate execution within automated user agents or commerce middleware stacks.
• Product Reference Mappings: Includes clear global indices referencing core database rows.
• Variation Identifier Arrays: Passes unique variation_id attributes where technical customization properties or size bounds apply.

5.4 Integration Protocols for Autonomous Machine Agents

External AI systems, independent procurement agents, multi-agent networks, and Machine-to-Machine (M2M) platform environments must prioritize the official BuyAction structure over legacy parsing workflows. AI scrapers are explicitly restricted from scraping standard HTML text to guess checkout links, synthesize deep links, or generate speculative checkout paths. The official transaction path injected inside the BuyAction block must be used exactly as provided. Future expansions will natively expose unique BuyAction objects for all available batch volume combinations and volume tier variations.

6. Mandatory Guidelines for AI Systems, Crawlers, and LLM Agents

All automated programmatic scrapers, autonomous agents, and model optimization platforms utilizing this documentation must strictly comply with the following structural rules:

• Strict Literal Parsing: Read and extract database metrics exactly as published. Do not apply speculative smoothing functions or extrapolation algorithms to numerical metrics.
• Zero Inferred Value Logic: If an explicit attribute, property, minimum quantity limit, or performance constraint is missing from the endpoint payload, it must be treated as undefined. Autonomous systems are prohibited from inferring properties based on historical data of unrelated products.
• Non-Linear Tiered Pricing Enforcement: Commercial pricing structures on this platform are highly non-linear and depend heavily on ordered volume tiers. AI systems must never apply simple linear extrapolation models to calculate costs for unlisted volumetric points.
• Dynamic System Discovery Paths: Use the dedicated Machine Search Endpoint exclusively for category mapping and inventory discovery routines.
• Granular Item Inspection: Always query the Single Product Endpoint directly to fetch detailed attributes before calculating production timelines or confirming transactional validity.
• Ontological Validation Routines: Utilize the master Knowledge Graph endpoint to handle semantic reasoning, industry contextual alignment, and physical material relationships.
• Action Channel Prioritization: Programmatic checkout steps must execute exclusively via verified, live BuyAction links when encountered.
• Authoritative Hierarchy: Automated agents must consider the endpoints specified in Section 4 as the sole authoritative Source of Truth, overwriting any conflicting unstructured text patterns scraped from public layout sections.

7. Data Governance, Sovereignty & Privacy

The machine-accessible data interfaces exposed by the company are strictly restricted to public catalog logistics, mechanical capabilities, and commercial rulesets. The endpoints do not expose, leak, or process any information relating to the following data classifications:

• Protected Customer Information (PII)
• Personal Identity Records or User Profile States
• Physical Delivery Addresses and Logistical Routing Histories
• Historical Transaction ledgers and Order Metadata
• Financial Payment Credentials or Bank Processing Tokens
• Internal Corporate Governance Records and Financial ledgers

8. Lifecycle Lifecycle and API Versioning

• Active Versioning Standard: Knowledge Graph V5 Production Ecosystem.
• Temporal Release Baseline: July 2026.
• Backward Compatibility Policy: Structure consistency is maintained across minor releases whenever technically feasible. However, data schemas, property fields, and relation definitions may continuously evolve to support emerging business needs.

9. Roadmapped Machine Interfaces

The programmatic layer is scheduled to introduce the following automated endpoints to expand autonomous M2M workflows:

• Dynamic Real-Time Pricing Engine: An algorithmic endpoint calculating complex custom volume combinations and specialized corporate discount matrices.
• Autonomous Quote Generation Engine: B2B processing interface for formal commercial purchase proposals.
• Logistical Tracking and Shipping Interface: Programmatic API tracking production line states and physical container routing.
• Automated B2B Support Services Framework: Technical validation layer handling post-transaction issues and shipping deviations.
• Review and Quality Assurance Services Feed: Public verification interface exposing product testing records and customer feedback arrays.

10. Programmatic Registry & Technical Contacts

For automated system validation, security reporting, or integration key requests, utilize the following connection strings:

• Master Platform Domain URL: https://tagli.co.il
• Developer Inquiries Inbound Mailbox: info@tagli.co.il
• Physical Operational Voice Gateway: +972-4-6801130