The intelligence platform that changes how TMF APIs work.

Introduction

The TMF API Intelligence Toolkit is an enterprise-grade, browser-local suite designed for telecom solution architects, integration engineers, and API managers working with TM Forum (TMF) Open API standards.

Managing TM Forum specifications involves navigating intricate nested object hierarchies, polymorphic sub-types (FVO/MVO), complex characteristics arrays, and multi-version schema rules across 23 core domain specifications. The Toolkit solves this by converting dense JSON payloads and OpenAPI schemas into interactive, inspectable topological graphs and providing instant client-side validation.

Platform Architecture

The application follows a decoupled, browser-native pipeline architecture:

Core Architectural Components

Privacy & Security Model

Core Concepts & Terminology

Interactive Visualizer Engine

The Visualizer engine renders complex multi-level JSON payloads into visual topology graphs. It handles 23 distinct TMF domain specifications, automatically grouping items, resolving relationships, and highlighting status lifecycles.

Key Features

• Interactive Node Graph: Pan, zoom, and collapse/expand sub-trees for large order payloads containing hundreds of items.
• Polymorphic Type Badges: Color-coded badges for @type, @baseType, @schemaLocation, and action states (add, modify, delete, noChange).
• Characteristic Extraction: Automatically parses name/value name-value pair arrays into clean property cards.

Payload Inspection & Deep-Drill

Clicking any node in the topology graph opens the Deep Inspection Drawer:

{
  "id": "PO-10492",
  "category": "B2B Fiber Service",
  "productOrderItem": [
    {
      "id": "1",
      "action": "add",
      "state": "acknowledged",
      "product": {
        "name": "Gigabit Enterprise Fiber",
        "productCharacteristic": [
          { "name": "DownloadSpeed", "value": "1000Mbps" },
          { "name": "VLAN_ID", "value": "4092" }
        ]
      }
    }
  ]
}

Laser Focus Mode

Designed for architecture walkthroughs and executive reviews: clicking Laser Mode isolates the selected node and its direct ancestor/descendant relationships while dimming background clutter.

Full-Height PNG Export

Clicking Export PNG renders the entire canvas at high resolution, capturing full document height and node connections for inclusion in architecture design documents (ADDs) or client reports.

Schema Validator Engine

The Schema Validator evaluates JSON payloads against official published TM Forum OpenAPI specification contracts.

OpenAPI v4 vs v5 Specification Handling

Schema Diff Analyzer

The Diff Analyzer compares two OpenAPI schema versions (e.g. TMF622 v4.0.0 vs v5.0.0) or two custom JSON payloads to identify structural changes, property type mutations, and breaking contract modifications.

Breaking Change Detection Rules

• Removed Properties: Highlights mandatory fields removed in newer specification versions.
• Type Mutations: Detects fields changed from string to object or array.
• Enum Constraint Tightening: Surfaces restricted enum value sets.

Spectra AI Engine

Spectra AI is an advanced multi-engine AI co-pilot designed specifically for TM Forum Open API architecture, payload inspection, and schema validation.

Ollama & LM Studio Local Engine Integration

In addition to browser-local WebGPU, Spectra AI supports direct client-side integration with host-level LLM servers including Ollama and LM Studio. This allows architects to run massive local models (7B to 70B+, DeepSeek-R1, Qwen2.5-Coder, Llama 3.3) directly on their workstation with zero API keys and zero cloud cost.

1. Ollama Setup & CORS Configuration

By default, browsers enforce Cross-Origin Resource Sharing (CORS) security when a hosted web page accesses `localhost`. Launch Ollama with the OLLAMA_ORIGINS environment variable to grant browser access:

# Windows Command Prompt (cmd)
set OLLAMA_ORIGINS=* && ollama serve

# Windows PowerShell
$env:OLLAMA_ORIGINS="*" ; ollama serve

# macOS / Linux Terminal
OLLAMA_ORIGINS="*" ollama serve

Spectra AI queries http://localhost:11434/api/tags to automatically detect all pulled models on your machine (e.g. qwen2.5:7b, deepseek-r1:8b, llama3.2) and populates them in the model picker dropdown.

2. LM Studio Setup & CORS Configuration

1. Launch LM Studio and load your desired model (e.g., Qwen 2.5 Coder 14B or DeepSeek R1 Distill).
2. Open the Local Server / Developer tab (↔) in LM Studio.
3. Set Port to 1234 and check the box for Enable CORS.
4. Click Start Server.

Spectra AI automatically connects to http://localhost:1234/v1/models to detect the active model loaded in LM Studio.

WebGPU Model Shard Auto-Retry & Network Resilience

When loading local WebGPU models (e.g. 2B or 4B Qwen 3.5), model parameter files (~1.2 GB to ~2.4 GB) are downloaded from HuggingFace CDN in sharded binary blocks (e.g. params_shard_20.bin).

Because downloaded shards are cached incrementally using the browser's native Cache Storage API, an auto-retry resumes instantly from the exact percentage (e.g. 82%) where the network dropped, ensuring uninterrupted completion without losing progress.

Browser-Local Knowledge Base & Vector RAG

The Knowledge Base allows you to upload custom PDF user guides, architecture standards, and domain documentation directly into your browser to power Retrieval-Augmented Generation (RAG).

How Vector RAG Pipeline Works

1. Document Ingestion: Upload PDF, MD, TXT, or JSON files into the Knowledge Base modal.
2. Client-Side Text Extraction: pdfjs-dist extracts page-by-page text streams in-memory.
3. Semantic Text Chunker: Text is split into ~500-token chunks with 50-token sliding window context overlap.
4. Vector Embeddings: Feature vectors are computed locally for each chunk.
5. IndexedDB Storage: Chunks and metadata persist in origin-isolated IndexedDB (tmf_vector_rag_v1).
6. Augmented Prompt Injection: On every user query, cosine similarity retrieves top matching chunks and injects them into Qwen 3.5 context with exact page citations.

23 Supported TM Forum Open API Specifications

IndexedDB Storage Schemas (tmf_vector_rag_v1)

Database: tmf_vector_rag_v1 (Version 2)
  ├── schema_chunks        [Key: id, Index: apiId, api_ver]
  ├── indexing_registry    [Key: key]
  ├── custom_doc_chunks    [Key: id, Index: docId, apiId]
  └── custom_doc_registry  [Key: docId]

Troubleshooting & Browser Requirements