nchhaya commited on
Commit
3d4af92
·
verified ·
1 Parent(s): d62609c

Reposition org card: HyperAPI for enterprise documents, datasets suite, agents, production results

Browse files
Files changed (1) hide show
  1. README.md +59 -44
README.md CHANGED
@@ -6,78 +6,93 @@ website: https://www.hyperbots.com
6
  type: company
7
  location: New York, NY & Bengaluru, India
8
  ---
9
- # Hyperbots: Agentic AI for Finance & Accounting
10
 
11
- Hyperbots builds domain-specific AI for the Office of the CFO. Our models are trained to read financial documents the way a finance professional does: the line items on an invoice, the terms in a contract, the structure of a general ledger.
12
 
13
- ---
 
 
 
 
 
 
14
 
15
- ## ⚡ HyperAPI: financial document intelligence for developers
 
 
16
 
17
- **[HyperAPI](https://apis.hyperbots.com)** gives developers the same document models that run the Hyperbots platform. You upload a PDF, scan or image and get back validated JSON. HyperAPI is built for backend services, dashboards and AI agents.
18
 
19
- | Endpoint | What it does | Price |
20
  |---|---|---|
21
- | `POST /v1/parse` | OCR to clean text and markdown, with layout regions, line and word boxes, and tables | from **$0.001**/page |
22
- | `POST /v1/extract` | Typed fields and line items from invoices and receipts, or any document with your own JSON schema | $0.05/page |
23
- | `POST /v1/classify` | Document type with a confidence score and ranked candidates; custom classes supported | $0.05/page |
24
- | `POST /v1/split` | Breaks a bundled PDF into separate documents, with page ranges and labels | $0.05/page |
25
- | `POST /v1/redact` | Masks PII with black boxes or replaces it with synthetic values | $0.05/page |
26
 
27
- **Start free:** 30 pages/month with no credit card. Paid plans start at $1/month. See [pricing](https://apis.hyperbots.com/pricing).
28
 
29
  ```bash
30
  pip install "hyperapi-sdk @ git+https://github.com/hyprbots/hyperapi-sdk.git"
31
- export HYPERAPI_KEY=hk_test_... # https://apis.hyperbots.com/dashboard/api-keys
32
  ```
33
 
34
  ```python
35
  from hyperapi import HyperAPIClient
36
 
37
  with HyperAPIClient() as client:
38
- invoice = client.process("invoice.pdf") # parse + extract in one call
39
- print(invoice["result"])
 
 
40
  ```
41
 
42
- <p>
43
- <a href="https://huggingface.co/spaces/hyperbots-inc/HyperAPI"><b>🧾 HyperAPI Space</b></a> ·
44
- <a href="https://apis.hyperbots.com/docs"><b>📚 Docs</b></a> ·
45
- <a href="https://apis.hyperbots.com/sandbox"><b>🧪 Sandbox</b></a> ·
46
- <a href="https://github.com/hyprbots/hyperapi-sdk"><b>🐍 Python SDK</b></a> ·
47
- <a href="https://apis.hyperbots.com/blogs"><b>✍️ Engineering blog</b></a>
48
- </p>
49
 
50
- **Why developers choose it**
51
- - **Built for finance documents:** invoices, receipts, bank statements, POs, contracts and W-9s, in PDF, image and Office formats.
52
- - **Composable:** upload once, then chain Parse → Classify → Split → Extract → Redact using the same `document_key`.
53
- - **Production features:** sync and async clients, async jobs (`X-Async: true`), batches, webhooks and 24-hour result caching. Cached results are not billed.
54
- - **Private by default:** files are processed in memory and deleted after each run unless you turn on retention, and documents are not used for training.
55
 
56
- ---
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
57
 
58
- ## 📦 On Hugging Face
59
 
60
- | | |
61
- |---|---|
62
- | 🧾 [**HyperAPI** (Space)](https://huggingface.co/spaces/hyperbots-inc/HyperAPI) | Developer landing page with endpoints, quickstart, REST examples and pricing |
63
- | 📊 [**synthetic-invoice-extraction** (Dataset)](https://huggingface.co/datasets/hyperbots-inc/synthetic-invoice-extraction) | 200 fully synthetic invoices (HTML, OCR text and gold labels in the HyperAPI Extract format), plus a script to benchmark extraction |
64
 
65
- ## 🔬 Research
66
 
67
- - **DocuLite: Breaking the annotation barrier.** A framework for adapting LLMs and VLMs to invoice extraction without human-annotated training data. It uses synthetic text (InvoicePy) and HTML templates (TemplatePy) to train small models, an OpenChat-7B and an InternVL-2-8B. Accepted at the 4th Deployable AI Workshop, AAAI 2026. [Read more](https://www.hyperbots.com/research/breaking-the-annotation-barrier-with-doculite)
 
 
 
68
 
69
- ## 🛠 The technology behind it
70
 
71
- - **HyperLM:** a language model trained on financial data, including invoices, POs, contracts and GL structures. It is the core model behind HyperAPI.
72
- - **Vision-language models:** multimodal models that read low-quality scans, handwriting and dense tables.
73
- - **Agentic reasoning:** a multi-agent layer that handles validation, anomaly detection and ERP write-backs, with full audit trails.
74
 
75
- On the Hyperbots platform, these models also run GL coding, 2- and 3-way PO matching, accruals, cash-flow forecasting and sales-tax verification. They connect natively to `NetSuite`, `SAP S/4HANA`, `Sage Intacct`, `Microsoft Dynamics 365` and `QuickBooks`.
76
 
77
- ## 🔒 Security & trust
78
 
79
- SOC 1 Type 2, SOC 2 Type 2 and ISO 27001 certified. A redaction layer removes PII and PHI before any model sees the data, and a human-in-the-loop workflow shows reviewers the reasoning behind each AI decision.
80
 
81
- ## 🤝 Connect
82
 
83
- [HyperAPI](https://apis.hyperbots.com) · [Website](https://www.hyperbots.com) · [Blog](https://blog.hyperbots.com/) · [LinkedIn](https://www.linkedin.com/company/hyperbots-inc) · [X](https://x.com/hyprbots) · [hyperapi-support@hyperbots.com](mailto:hyperapi-support@hyperbots.com)
 
6
  type: company
7
  location: New York, NY & Bengaluru, India
8
  ---
9
+ # HyperAPI by Hyperbots: document intelligence APIs for enterprise solutions
10
 
11
+ **One API to parse, extract, classify, split and redact any enterprise document**, from invoices, statements and POs to contracts, KYC packs, HR files, claims, lab reports and email. It returns schema-validated JSON with a confidence score and a source bounding box for every field. HyperAPI runs inside your own applications, and AI agents can call the same operations over MCP.
12
 
13
+ <p>
14
+ <a href="https://apis.hyperbots.com/dashboard/register"><b>🔑 Get a free API key</b></a> ·
15
+ <a href="https://apis.hyperbots.com/sandbox"><b>🧪 Try the sandbox</b></a> ·
16
+ <a href="https://apis.hyperbots.com/docs"><b>📚 Docs</b></a> ·
17
+ <a href="https://huggingface.co/spaces/hyperbots-inc/HyperAPI"><b>🧾 HyperAPI Space</b></a> ·
18
+ <a href="https://huggingface.co/collections/hyperbots-inc/hyperapi-enterprise-document-intelligence-6af6c8764547d894d4e1a3c4"><b>🗂 Collection</b></a>
19
+ </p>
20
 
21
+ | 99.8% | < 2 s | 99.9% | 1M+ pages |
22
+ |:---:|:---:|:---:|:---:|
23
+ | extraction accuracy (reported) | average latency | uptime SLA | digitised in 10 days for one client |
24
 
25
+ ## Five operations, one endpoint
26
 
27
+ | Operation | What you get | Typical latency |
28
  |---|---|---|
29
+ | **Parse** `POST /v1/parse` | Layout-aware text and Markdown, reading order, tables, word boxes with fonts. Handles scans, photos, multi-column pages and degraded PDFs | ~1 s |
30
+ | **Extract** `POST /v1/extract` | Typed fields and any number of line items. Use the built-in financial format or pass a JSON schema for any document type | ~2 s |
31
+ | **Classify** `POST /v1/classify` | Document type with a confidence score. 67 types out of the box, and it flags types outside the set | ~0.8 s |
32
+ | **Split** `POST /v1/split` | Document boundaries in a bundled file, detected from content rather than page counts | ~0.5 s |
33
+ | **Redact** `POST /v1/redact` | Names, account numbers, IBANs, ID numbers and other PII masked or replaced with synthetic values | ~3 s |
34
 
35
+ Every response includes per-field confidence, source boxes, document-level checks (for example, line items must add up to the total), and `request_id`, `model_used`, duration and page count for audit. You can run calls sync, async (`X-Async`) or through the Batch API with webhooks. Results are cached for 24 hours, and cached re-runs are not billed.
36
 
37
  ```bash
38
  pip install "hyperapi-sdk @ git+https://github.com/hyprbots/hyperapi-sdk.git"
39
+ export HYPERAPI_KEY=hk_test_... # free tier: 30 pages/month, no credit card
40
  ```
41
 
42
  ```python
43
  from hyperapi import HyperAPIClient
44
 
45
  with HyperAPIClient() as client:
46
+ doc = client.process("packet.pdf") # parse + extract in one call
47
+ kind = client.classify("packet.pdf") # what is it?
48
+ parts = client.split("packet.pdf") # where does each document start and end?
49
+ safe = client.redact("packet.pdf", mode="deidentify")
50
  ```
51
 
52
+ ## What it reads
 
 
 
 
 
 
53
 
54
+ - **Formats:** native, scanned and mixed PDFs (including forms and files of several hundred pages); PNG, JPEG, TIFF, HEIC and other images; Word, Excel and PowerPoint, with workbooks parsed natively; and email (EML, MSG).
55
+ - **Document types:** a 100-type reference taxonomy in 15 categories, covering payables, statements, payments, procurement, logistics, HR, onboarding and compliance, financial statements, credit and lending, healthcare, legal, and operations.
56
+ - **Hard pages:** skewed, rotated and shadowed scans, stamps and seals, handwriting, dense multi-column layouts, and tables with hundreds of rows. Text in more than 100 languages, including mixed-language documents.
 
 
57
 
58
+ ## 📊 Open synthetic datasets for document AI
59
+
60
+ These are fully synthetic, labelled in HyperAPI's output formats, and free to use under CC BY 4.0. Use them to benchmark HyperAPI or any other document pipeline.
61
+
62
+ | Dataset | Task | Size |
63
+ |---|---|---|
64
+ | [synthetic-enterprise-doc-extraction](https://huggingface.co/datasets/hyperbots-inc/synthetic-enterprise-doc-extraction) | Field extraction for 31 document types, with HyperAPI schemas | 372 docs |
65
+ | [synthetic-doc-parsing](https://huggingface.co/datasets/hyperbots-inc/synthetic-doc-parsing) | Layout blocks, reading order, tables, Markdown | 100 pages |
66
+ | [synthetic-doc-classification](https://huggingface.co/datasets/hyperbots-inc/synthetic-doc-classification) | 31-class classification on the Hyperbots taxonomy | 248 pages |
67
+ | [synthetic-doc-splitting](https://huggingface.co/datasets/hyperbots-inc/synthetic-doc-splitting) | Boundaries in multi-document bundles | 200 bundles |
68
+ | [synthetic-pii-redaction](https://huggingface.co/datasets/hyperbots-inc/synthetic-pii-redaction) | PII spans and boxes, redacted and de-identified text | 100 pages |
69
+ | [synthetic-handwritten-forms](https://huggingface.co/datasets/hyperbots-inc/synthetic-handwritten-forms) | Handwritten fields, checkboxes, signatures | 100 pages |
70
+ | [synthetic-stamped-documents](https://huggingface.co/datasets/hyperbots-inc/synthetic-stamped-documents) | Stamps and seals over printed text | 100 pages |
71
+ | [synthetic-bad-scans](https://huggingface.co/datasets/hyperbots-inc/synthetic-bad-scans) | Skew, blur, noise and compression at 3 severities | 120 pages |
72
+ | [synthetic-dense-documents](https://huggingface.co/datasets/hyperbots-inc/synthetic-dense-documents) | Dense directories, legal terms, long tables | 60 pages |
73
+ | [synthetic-invoice-extraction](https://huggingface.co/datasets/hyperbots-inc/synthetic-invoice-extraction) | Invoice fields and line items, plus a scorer | 200 docs |
74
 
75
+ ## Built for agents
76
 
77
+ HyperAPI exposes its five operations over **MCP**, so coding and workflow agents (Claude Code, Cursor, GitHub Copilot, Codex, OpenCode and any other MCP client) can read documents reliably. With **Agent Builder (beta)**, you give an agent documents and a goal, for example *"reconcile this month's vendor invoices against open POs, flag anything over $500 and mask bank details"*. The agent picks the operations, runs them with guardrails and returns one structured result with a citation for every field.
 
 
 
78
 
79
+ ## In production
80
 
81
+ - **Invoice processing at scale:** a US marketing-services enterprise with 21 entities on SAP and NetSuite processes more than 1M invoices a year through an agent built on HyperAPI. It reports 99%+ extraction accuracy, 80% less manual AP work and 40% lower AP cost.
82
+ - **$10B manufacturer:** more than 300K invoices a year posted into SAP ECC, with a 40% reduction in outsourced (KPO) staff.
83
+ - **Dense archive digitisation:** more than 1M multi-column directory pages parsed into structured records within a fixed 10-day window.
84
+ - **Document reconstruction pilot:** one `parse` call per file across vector PDF, scanned PDF, XPS, Outlook email and Excel, with 97% median text recall on the client's own files.
85
 
86
+ ## The models
87
 
88
+ HyperAPI runs on **HyperLM**, Hyperbots' mixture-of-experts document stack. It combines small, vision and large language models with multimodal encoders for layout, text and image, and decoders specialised by task and field. HyperLM is trained on **35M+ real-world documents and 0.5B synthetic samples** and retrained continuously. Related research: [**DocuLite**](https://www.hyperbots.com/research/breaking-the-annotation-barrier-with-doculite) (4th Deployable AI Workshop, AAAI 2026) shows how to adapt small LLMs and VLMs to invoice extraction with synthetic data instead of human annotation.
 
 
89
 
90
+ ## Security and enterprise deployment
91
 
92
+ ISO 27001, SOC 1 Type 2 and SOC 2 Type 2. Documents are processed in memory and deleted when the run completes unless you enable retention, and customer data is never used for training. Redaction can run before a document reaches any model. BYOK, IP allow-listing, SSO (Okta, Entra ID, Google Workspace, OneLogin), role-based access and full audit trails are available. Enterprise evaluations run as a POC on your own sample documents, scored field by field against an answer key.
93
 
94
+ ## About Hyperbots
95
 
96
+ Hyperbots builds AI agents for finance and accounting (procure-to-pay, order-to-cash and the CFO office). HyperAPI is the same document intelligence that runs inside those agents, offered as APIs. The team has 150+ people, including 50 machine-learning scientists, with 50+ patents and 100+ publications.
97
 
98
+ [apis.hyperbots.com](https://apis.hyperbots.com) · [hyperbots.com](https://www.hyperbots.com) · [GitHub SDK](https://github.com/hyprbots/hyperapi-sdk) · [LinkedIn](https://www.linkedin.com/company/hyperbots-inc) · [X](https://x.com/hyprbots) · [hyperapi-support@hyperbots.com](mailto:hyperapi-support@hyperbots.com) · [research@hyperbots.com](mailto:research@hyperbots.com)