The Agentic
Document intelligence API.

Vision parsing, structured extraction, and conversational agents—one API. Upload anything, get validated data back.

99.2%

Extraction Accuracy

<3s

Avg Latency

10GB

Max Payload

extract.ts
const response = await fetch('https://api.getneji.com/api/extract', {
  method: 'POST',
  headers: {
    'Authorization': `Bearer ${process.env.NEJI_API_KEY}`,
    'Content-Type': 'application/json'
  },
  body: JSON.stringify({
    projectId: 'proj_invoices',
    fileId: 'file_receipt_001',  // uploaded via POST /api/files/upload
    schema: {
      type: 'object',
      properties: {
        vendor: { type: 'string', description: 'Merchant name from header' },
        date: { type: 'string', format: 'date' },
        lineItems: {
          type: 'array',
          items: {
            type: 'object',
            properties: {
              description: { type: 'string' },
              amount: { type: 'number' }
            }
          }
        },
        total: { type: 'number' }
      },
      required: ['vendor', 'total']
    }
  })
});

const { data, confidence, validation } = await response.json();
// data: { vendor: "Whole Foods", total: 84.32, lineItems: [...] }
// confidence: { vendor: { level: "high", source: "Page 1, header" }, ... }

PDFs, images, scans—any document.

One API. No pipeline.

Replace OCR, extraction, and Q&A with one endpoint.

Vision API

Multimodal parsing for images—receipts, charts, handwriting, scanned docs. No OCR preprocessing layer.

Schema Extraction

Pass a JSON Schema, get validated output. Confidence scores and source locations for every field.

Custom Agents

Define agents with system prompts and tool permissions. Upload Skills as markdown to encode domain logic.

Batch & Queue

Process thousands of documents async. Fair queuing, rate limiting, and webhook callbacks built in.

What you can build

InvoicesReceiptsContractsFormsChartsReports

Multimodal by default.

Images are parsed natively—no Tesseract, no preprocessing. Layout, typography, handwriting—all understood. Structured output returned.

Vision API docs
receipt_march_15.jpgprocessing

detected

receipt · retail

layout

structured

text_blocks

24

handwriting

false

Schema in, JSON out.

Define your schema once with JSON Schema. Every extraction returns typed data plus confidence metadata: level (high/medium/low), source location, and extraction notes.

Extraction docs
vendorhigh
"Acme Corp"
amounthigh
1250.00
dueDatemedium
"2026-02-15"

Agents with memory.

Multi-turn sessions that persist context. Configure agents with custom prompts and tools. Upload Skills as markdown to encode procedures.

Agents docs

invoice-processor

3 skills loaded

ReadExtractValidate

prompt: "Extract line items, validate totals..."

Why not Tesseract + regex?

Skip the OCR → parser → validator stack.

Feature
Neji
Others
Native vision (no OCR layer)
Per-field confidence scores
Custom agents + skills
Multi-turn chat sessions
partial
Async batch processing
JSON Schema validation

Core endpoints

POST /files/upload

Stream PDFs, images, or text. Extraction and indexing handled async.

POST /extract

Pass JSON Schema with field descriptions. Get validated output + confidence.

POST /chat/:projectId

Ask follow-up questions. Context persists across turns.

Pricing

Pay per request. No seat licenses.

Free

100 req/mo

For prototyping

$29

5K req/mo

For small teams

Popular

$99

50K req/mo

For production

Custom

Unlimited

Volume discounts

Ship document features,
not infrastructure.

Free tier. No credit card.