> 🚧 *Portal docs are evolving — more APIs and guides coming soon.* # Welcome to LDX hub **LDX hub** is a document AI gateway — built on 5 years of production infrastructure and 100+ integrated AI vendors. Originally developed at [Kawamura International](https://ldxlab.io), a Tokyo-based language services company with 40 years of enterprise expertise. The APIs in this portal are ready-to-use endpoints that wrap the core LDX hub engine. No infrastructure setup required — bring your API key and start processing. --- ## Try it in 60 seconds Get a free API key, then run two `curl` commands — the second pipes the first's `job_id` and waits for the result. **1. Get your API key** — [Sign up free](/signup) and grab your API key from the dashboard. No credit card required, 25,000 credits included. **2. Submit and fetch** — extract `name`, `email`, and `intent` from a message: *Requires `jq` for parsing JSON in the shell. Install via your package manager (`brew`, `winget`, `apt`, etc.) — or just copy the `job_id` manually from the response.* ```bash # Submit the job and capture the job_id JOB_ID=$(curl -s -X POST https://gw.ldxhub.io/structflow/jobs \ -H "Authorization: Bearer YOUR_API_KEY" \ -H "Content-Type: application/json" \ -d '{ "model": "google/gemini-3.6-flash", "system_prompt": "Extract the customer name, email, and intent from the message.", "example_output": {"name": "John Doe", "email": "john@example.com", "intent": "quote_request"}, "inputs": [ {"id": "1", "data": {"message": "Hi, I am Sarah Chen (sarah@acme.io). We need a quote for 50 seats."}} ] }' | jq -r .job_id) # Fetch the result — the server holds the connection until the job completes (up to 10s) curl "https://gw.ldxhub.io/structflow/jobs/$JOB_ID?wait=10" \ -H "Authorization: Bearer YOUR_API_KEY" ``` Returns clean, structured JSON: ```json { "status": "completed", "results": [ { "id": "1", "status": "succeeded", "output": { "name": "Sarah Chen", "email": "sarah@acme.io", "intent": "Request a quote for 50 seats" } } ] } ``` That's it. Two calls, piped — submit, then wait. The `wait=10` parameter holds the connection open for up to 10 seconds, so the result returns as soon as the job is done. For longer-running jobs, just call the same `GET` again — each call holds for another 10 seconds. Now try it with your own data, your own schema, your own scale. --- ## Built for scale, by design That same endpoint accepts thousands of inputs in a single call. Drop in 10,000 records — StructFlow runs them in parallel server-side and returns results when the job completes. ```bash curl -X POST https://gw.ldxhub.io/structflow/jobs \ -H "Authorization: Bearer YOUR_API_KEY" \ -H "Content-Type: application/json" \ -d '{ "model": "google/gemini-3.6-flash", "system_prompt": "Extract the customer name, email, and intent from the message.", "example_output": {"name": "John Doe", "email": "john@example.com", "intent": "quote_request"}, "inputs": [ {"id": "1", "data": {"message": "Hi, I am Sarah Chen (sarah@acme.io). We need a quote for 50 seats."}}, {"id": "2", "data": {"message": "Hey — Mike from BetaCorp here, mike@beta.io. Can you send pricing?"}}, {"id": "3", "data": {"message": "Need to renew our contract. — Lisa Park, lisa@gamma.co"}} ] }' ``` For larger workloads, upload a JSONL file and pass `file_id` instead of `inputs`. No rate limit juggling, no client-side concurrency, no retries to write yourself. The async-first design is what makes this scale possible — every endpoint in LDX hub works the same way. --- ## What You Can Do Here | Service | What it does | |---|---| | **StructFlow** | Unstructured text → structured JSON | | **AnalyzeDoc** | PDF / images → structured JSON (vision AI) | | **RefineLoop** | Machine translation → refined translation (XLIFF) | | **RenderOCR** | PDF / images → editable Office files (with OCR) | | **CastDoc** | Text-based PDF → editable Office files (no OCR) | | **ExtractDoc** | Documents → plain text (preprocessing for AI pipelines) | ### StructFlow **Turn unstructured text into structured data. At scale.** Free-form text — medical records, contracts, reviews, HR notes — is everywhere. StructFlow takes that text, applies your schema and instructions, and returns clean, validated JSON. Fast. In parallel. With output structure guaranteed by Structured Outputs technology. - Works with any major AI engine (Claude, Gemini, GPT, Amazon Nova, Grok, and more) - Define your output schema from a single sample JSON line - Built for batch: drop a JSONL file, get a JSONL file back - 5 years of production-hardened parallel processing under the hood → [Learn more about StructFlow](https://ldxlab.io/ldxhub/structflow) --- ### AnalyzeDoc **Let vision AI read your documents — and hand back structured data.** Some documents resist plain-text parsing — invoices with intricate layouts, scanned forms, receipts, visual tables. AnalyzeDoc uses vision-capable AI models to *look* at PDFs and images the way a person would, then returns clean JSON following your schema. No separate OCR step, no layout heuristics — the model reads and structures in a single pass. - Works with vision models from OpenAI, Azure, Google, Anthropic, and Amazon - Define your output schema from a single sample JSON object — just like StructFlow - Input: PDF, JPEG, or PNG — the format is detected automatically - Higher-resolution variants available for dense layouts and fine print, at a higher per-page cost --- ### RefineLoop **Machine translation quality, refined until it converges.** RefineLoop takes MT output in XLIFF format and runs it through an AI review loop — checking for mistranslations, omissions, and additions across three axes. It keeps revising until the AI agrees with itself. That's the convergence. - Drops into any existing translation pipeline as a single XLIFF step - Compatible with XTM, memoQ, Trados, and other CAT tools - Revision history recorded in `alt-trans` tags — fully auditable - Powered by the same parallel processing engine as StructFlow → [Learn more about RefineLoop](https://ldxlab.io/ldxhub/refineloop) --- ### RenderOCR **Scanned documents, converted to editable Office files — with layout intact.** PDFs and images sitting in your archive aren't searchable, editable, or pipeline-ready. RenderOCR runs high-fidelity OCR across your documents and returns Word, Excel, or PowerPoint files that preserve the original layout and formatting — not just raw text dumps. - Supports PDF, TIFF, JPEG, PNG, and BMP as input - Output to DOCX, XLSX, or PPTX — your choice - 120+ languages supported, including Japanese, Chinese, Korean, Arabic, and more - Powered by KI OCR, a battle-tested enterprise OCR engine --- ### CastDoc **PDF files, converted to editable Office formats — without OCR.** Already have a text-based PDF? CastDoc converts it directly to Word, Excel, or PowerPoint while preserving the original layout and formatting — no optical recognition required. Faster and more accurate than OCR for born-digital documents. - Input: text-based PDF (not scanned) - Output to DOCX, XLSX, or PPTX — your choice - Powered by KI Cast - Ideal for contracts, reports, and any PDF created from digital sources --- ### ExtractDoc **Documents to plain text — fast, deterministic, and pipeline-ready.** Need just the text from a document? ExtractDoc pulls plain text from PDFs and Office files in reading order, with no AI overhead and no layout assumptions. Designed as the entry point for StructFlow and other downstream AI pipelines. - Supports PDF, DOCX, XLSX, and PPTX as input - Output as plain text or as JSONL — directly compatible with StructFlow's `file_id` input - No AI billing, no language settings, no parameters to tune - Built to chain: pair with StructFlow for extraction, RefineLoop for translation review --- ## Getting Started [Sign up free](/signup) to get your API key instantly — no credit card required, 25,000 credits included. Then head to the [API Reference](/api) to explore all available endpoints, request/response formats, and authentication details. --- ## Credits & Pricing LDX hub uses a credit-based billing system. Credits are consumed based on the engine you choose and the volume of data processed. → [Credits & Pricing](/credits-and-pricing) --- ## Use Anywhere LDX hub works as a backend for your existing tools: - **Dify** — [LDX hub plugin](https://marketplace.dify.ai/plugins/ldxhub-io/ldxhub) brings StructFlow and RefineLoop into your Dify workflows - **n8n** — [n8n-nodes-ldxhub](https://www.npmjs.com/package/n8n-nodes-ldxhub) exposes all 6 services in n8n automations - **Claude Desktop / MCP-compatible AI** — see [MCP Setup](/mcp-setup) below - **Direct API** — start with the [curl examples](#try-it-in-60-seconds) above --- ## Use with AI Assistants (MCP) LDX hub supports the [Model Context Protocol (MCP)](https://modelcontextprotocol.io), allowing you to use StructFlow, AnalyzeDoc, RefineLoop, RenderOCR, CastDoc, and ExtractDoc directly from AI assistants like Claude Desktop. → [MCP Setup Guide](/mcp-setup)