One API to parse, extract, classify, split and redact any enterprise document: invoices and statements, POs and bills of lading, contracts and NDAs, KYC packs, HR files, claims, lab reports and email. It returns schema-validated JSON with a confidence score and a source box for every field. You call it from your own applications, and an MCP server for AI agents is available on request.
Upload a file once and get a document_key, valid for 24 hours. Then run any operation against it, in any order. Each operation reads the page, interprets layout and content, and returns structured output that's ready for the next system.
POST /v1/parseLayout-aware text and Markdown with reading order, tables, labeled regions and word boxes with fonts. Handles scans, photos, multi-column layouts and low-quality PDFs.
POST /v1/extractTyped fields and any number of line items. Use the built-in financial format or send a JSON schema for any document type. The advanced extractor detects the type automatically.
POST /v1/classifyDocument type with a confidence score and ranked candidates. 67 types out of the box, flags types outside the set, and supports custom classes.
POST /v1/splitFinds where each document in a bundle starts and ends, from content rather than page counts. Trained on more than 1M page streams.
POST /v1/redactMasks names, account numbers, IBANs, IDs and other PII with black boxes or synthetic values. Run it first to keep PII out of LLM prompts.
Included with every call: per-field confidence and source boxes, document-level checks such as line items adding up to the total, request_id / model_used / duration / page count for audit, and sync, async (X-Async) or Batch API with webhooks. Cached re-runs within 24 hours are not billed.
No per-format templates. HyperAPI uses a 100-type reference taxonomy in 15 categories for classification and routing.
Get a free key (30 pages a month, no credit card), set HYPERAPI_KEY, and follow the quickstart to use the Python SDK or the REST API. An MCP server for agents is available on request.
from hyperapi import HyperAPIClient with HyperAPIClient() as client: doc = client.process("packet.pdf") # parse + extract in one call kind = client.classify("packet.pdf") # label + confidence parts = client.split("packet.pdf") # start/end page per document safe = client.redact("packet.pdf", mode="deidentify")
# any document, your schema (REST) curl -X POST "https://apis.hyperbots.com/v1/extract?category=non_financial" \ -H "X-API-Key: $HYPERAPI_KEY" \ -F "document_key=550e8400-e29b-41d4-a716-446655440000" \ -F 'schema={"policy_number":null,"claimant_name":null,"date_of_loss":null,"amount_claimed":null}'
Coding agents can follow a goal and write code, but they don't read unstructured documents reliably. HyperAPI's five operations can be exposed over MCP (available on request from hyperapi-support@hyperbots.com), so Claude Code, Cursor, GitHub Copilot, Codex, OpenCode or any MCP client can build the pipeline a goal needs. Agent Builder (beta) picks the operations, runs them with guardrails and returns one structured result with a citation for every field.
"Reconcile this month's vendor invoices against our open POs, flag anything over $500, and mask the bank details before it moves on."
Other workflows built from the same five operations include AP reconciliation, contract intake, statement digitisation, supplier onboarding and mixed-inbox triage.
Ten fully synthetic, CC BY 4.0 benchmarks, labelled in HyperAPI's output formats. Use them to evaluate HyperAPI or any document AI pipeline without sharing a single real document.
31 document types, 372 docs, gold fields plus the HyperAPI schema for each.
100 pages with layout blocks, reading order, tables and Markdown.
248 pages across 31 taxonomy classes, some stamped or degraded.
200 multi-document bundles with the correct boundaries.
100 pages with PII spans and boxes, plus redacted and de-identified text.
100 forms with handwritten fields, ticks, signatures and notes.
100 documents with stamps and seals over the text.
120 pages at 3 severities, with skew, blur, noise and compression.
60 high-resolution pages: directories, legal terms, long tables.
200 invoices with line items and a ready-made HyperAPI scorer.
Client names are withheld. All figures are as reported by Hyperbots.
A US marketing-services enterprise with 21 entities on SAP and NetSuite. An Invoice Processing Agent built on classify, split and extract reports 99%+ extraction accuracy, 80% less manual AP work and 40% lower AP cost.
More than 300K invoices a year posted into SAP ECC, with low-confidence fields routed to review and 40% fewer outsourced (KPO) staff.
An aged, dense, multi-column directory archive parsed with column boundaries and reading order intact, then extracted into structured records.
A document reconstruction pilot across vector PDF, scanned PDF, XPS, Outlook email and Excel, with 97% median text recall on the client's own files.
Start free. On paid plans, usage beyond the included pages is charged to prepaid credits at per-page rates.
| Plan | Price | Pages / month | Rate limit | Highlights |
|---|---|---|---|---|
| Free | $0 | 30 | 1 req/min | All five operations, fast parse, no credit card |
| Starter | $1 | 300 | 10 req/min | Email support |
| Plus | $25 | 10,000 | 30 req/min | Advanced parse, team workspace with 3 seats |
| Pro | $99 | 50,000 | 60 req/min | Priority routing, up to 10 seats |
| Custom / Enterprise | Talk to us | Volume | Custom throughput | BYOK, IP allow-listing, SSO, human review step, custom pipelines, forward-deployed engineers |
Current prices and promotions are on the pricing page.
HyperAPI runs on HyperLM, a mixture-of-experts document stack of small, vision and large language models, with multimodal encoders for layout, text and image and decoders specialised by task and field. It is trained on 35M+ real documents and 0.5B synthetic samples and retrained continuously.
Related research: DocuLite (4th Deployable AI Workshop, AAAI 2026) adapts small LLMs and VLMs to invoice extraction using synthetic data instead of human annotation.