Route any image or PDF through the optimal AI pipeline. Automatic fallback, cost-aware routing, and structured output. One API. No infrastructure to maintain.
Building vision pipelines means evaluating models, handling OCR fallbacks, writing retry logic, and watching costs spiral. Or you lock into one provider and hope it works on every input. There is a better way.
Docex maintains a live catalog of vision and OCR providers. For each job, it selects the optimal pipeline based on your input, budget, and latency requirements. When cheaper models suffice, you pay less.
If a provider fails or returns low confidence, Docex falls back to the next best option without any manual intervention. Your users never see an error they didn't cause.
Describe what you need in plain text. Docex returns validated JSON with confidence scores. No parsing headaches, no regex gymnastics, no brittle schemas to maintain.
We charge twice what the upstream provider charges us. No markup games, no opaque credits, no annual contracts. Refunds are issued if the actual cost comes in below the estimate.
Standard document extraction is where most tools stop. Docex starts there and keeps going.
IDs, licenses, bank statements, proof-of-address — extracted and validated in seconds.
Invoices, receipts, purchase orders — crumpled, low-light, any angle. Structured data out.
Email attachments, suspicious screenshots, scan reports — analyzed for threats and patterns.
Parties, clauses, signature detection, term extraction — without reading every page.
Shipping labels, waybills, manifests, damage photos — tracked and catalogued automatically.
Medical records, insurance claims, lab results, handwritten notes — digitized and structured.
We charge 2× what the upstream provider charges us. For most inputs, that means pennies per request.
Billed at 2× upstream cost. Dynamic routing means you pay less when cheaper models suffice.
$5 minimum top-up. Cancel anytime, no lock-in. No annual contract, no sales call. Refunds issued if actual cost is below the estimate.
Docex maintains a catalog of vision and OCR providers with their capabilities, costs, and latency profiles. For each job, the router classifies your input and selects the optimal pipeline. If a provider fails or returns low confidence, it falls back automatically.
PDF, JPEG, PNG, HEIC, and MP4 video frames. Docex handles conversion, preprocessing, and normalization automatically.
We bill at 2× the actual upstream provider cost. The DOCEX_MARKUP_MULTIPLIER env var controls this (default: 2). Jobs exceeding $1.00 require explicit confirmation. Refunds are issued if actual cost is below the estimate.
Docex manages provider authentication for you. You top up your Docex wallet and we handle the rest. No need to sign up for multiple AI providers or manage separate API keys.
Yes. The mock provider returns deterministic results for CI and testing at zero cost. Set DOCEX_ALLOW_MOCK_PROVIDER=1 to enable it.
Install the SDK (npm install docex), run docex setup, approve via GitHub, and call docex.run() with your file and prompt. Production-ready in under five minutes.
Paste this into your coding agent. It wires Docex into your product, scaffolds the endpoint for your stack, and runs a smoke test.