API Reference

IATE terminology (2.4M terms), neural translation (1,440+ language pairs), semantic search (20 Vectorize indexes), document annotation. All services run on Cloudflare Workers in EU jurisdiction.

Architecture

Pauhu® EU runs as a fleet of Cloudflare Workers in EU jurisdiction. Each service is a separate worker with its own domain. There is no single unified gateway URL — each API is accessed at its own worker endpoint.

All workers return JSON by default. CORS is enabled for pauhu.ai, pauhu.eu, and localhost development origins.

Authentication

API keys

Generate a self-service API key via the Pauhu search service. Keys follow the format pk_*. Each key is linked to a seat tier that determines data access entitlements.

Create an API key. Requires an email address. New keys start as live tier (activated by Stripe subscription).

curl -X POST https://staging.pauhu.eu/keys/generate \
  -H "Content-Type: application/json" \
  -d '{"email": "you@company.com"}'

Response:

{
  "api_key": "pk_...",
  "tier": "live",
  "entitlements": {
    "raw_feeds": true,
    "annotated_feeds": false,
    "training_export": false,
    "pauhu_ai": false,
    "search": true,
    "terminology": true,
    "translation": true,
    "rerank": true
  },
  "created_at": "2026-02-24T12:00:00.000Z"
}

Check current entitlements and burst limits. Requires Authorization: Bearer <api_key>.

{
  "tier": "live",
  "entitlements": {
    "raw_feeds": true,
    "annotated_feeds": false,
    "training_export": false,
    "pauhu_ai": false,
    "search": true,
    "terminology": true,
    "translation": true,
    "rerank": true
  },
  "burst_limit": "10 req/sec sustained, 50 req/sec peak",
  "active": true,
  "org_id": null
}

Request authentication

Include your API key as a Bearer token:

Authorization: Bearer pk_...

Without a key, requests run in trial mode: 3 requests/day (IP-based), search + terminology + translation only. No access to raw feeds, annotations, or training export.

Terminology API LIVE

Serves 2,456,445 IATE terms across 24 EU languages with exact lookup and semantic search (BGE-M3, 1024 dimensions).

Exact lookup

Exact match against IATE terminology database.

{
  "query": "data protection",
  "found": true,
  "count": 3,
  "results": [...],
  "source": "IATE",
  "stage": "LOOKUP"
}

Semantic search

BGE-M3 embedding search via Vectorize. Returns semantically similar terms ranked by cosine similarity.

curl -X POST https://staging.pauhu.eu/search \
  -H "Content-Type: application/json" \
  -d '{"query": "personal data processing", "lang": "en", "limit": 10}'

{
  "query": "personal data processing",
  "count": 10,
  "results": [...],
  "source": "IATE Pauhu Search",
  "stage": "MODEL"
}

Statistics

Term counts by language.

{
  "total": 2456445,
  "languages": 24,
  "byLanguage": [{"lang": "en", "count": 312847}, ...],
  "source": "IATE",
  "reliability": "4-star"
}

TBX export (ISO 30042)

Export terminology in TermBase eXchange format. Returns application/x-tbx+xml.

TMX export

Export translation pairs in Translation Memory eXchange format. Returns application/x-tmx+xml.

Batch export

Paginated export for embedding generation pipelines.

{
  "lang": "en",
  "offset": 0,
  "limit": 1000,
  "count": 1000,
  "hasMore": true,
  "nextOffset": 1000,
  "terms": [...],
  "embedding_model": "bge-m3-onnx"
}

Custom glossaries (tenant)

Upload a custom glossary (CSV or TBX). Terms are merged with IATE at lookup time, with tenant terms taking priority.

Merged lookup: tenant glossary first, then IATE fallback.

List all glossaries for a tenant.

Delete a tenant glossary.

Translation API LIVE

Helsinki-NLP OPUS-MT models at zero inference cost (browser ONNX). 1,440+ language pairs.

Translate text

6-stage translation cascade: KV cache → IATE terminology → rules engine → (reserved) → browser inference (OPUS-MT) → Vectorize semantic verification.

curl -X POST https://staging.pauhu.eu/translate \
  -H "Content-Type: application/json" \
  -d '{"text": "General Data Protection Regulation", "source_lang": "en", "target_lang": "fi"}'

{
  "source_lang": "en",
  "target_lang": "fi",
  "text": "General Data Protection Regulation",
  "translation": "Yleinen tietosuoja-asetus",
  "model_id": "Helsinki-NLP/opus-mt-en-fi"
}

Full cascade (verbose)

Returns all 6 cascade stages with timing and provenance for each step.

Batch segment translation

Translate an array of segments in a single request.

Supported languages

List all supported language codes and available pairs.

Search API LIVE

Fan-out semantic search across 20 indexes (BGE-M3, 1024 dimensions, cosine similarity). Powered by the Laine Algorithm.

Semantic search

Search across all 20 data source Vectorize indexes simultaneously.

{
  "query": "digital product passport ESPR",
  "limit": 20,
  "results": [
    {"product": "eurlex", "id": "32024R1781", "title": "...", "score": 0.89, "url": "..."},
    {"product": "commission", "id": "...", "title": "...", "score": 0.84, "url": "..."}
  ]
}

Instant answers

Knowledge panels from IATE, EUR-Lex, and Wikidata. Returns structured answer snippets.

Web proxy

CORS proxy for institutional search sources. Normalizes results into a common schema.

Cross-language siblings

Find all language versions of a EUR-Lex document by CELEX number.

Semantic reranking

Rerank a set of results using BGE-M3 cross-encoder scoring.

DLC packs (browser models)

Signed manifest of downloadable model packs with Ed25519 signatures and SHA-256 checksums.

Core DLC pack: ONNX models and terminology for browser-native inference.

Delta pack: incremental updates since last core download.

Annotation API LIVE

Classifies documents with topic annotations, deontic modalities, and language detection. Split across two service instances (A–E and E–W) for the 20 data sources.

Annotate document

Full annotation pipeline: language detection, topic classification, deontic modality (obligation/prohibition/permission/exemption), word count, product-specific metadata.

curl -X POST https://annotate.pauhu.eu/annotate \
  -H "Content-Type: application/json" \
  -d '{"text": "Member States shall ensure...", "product": "eurlex"}'

{
  "original_path": "...",
  "organized_path": "...",
  "language": "en",
  "deontic": {"modality": "obligation", "confidence": 0.95},
  "product": "eurlex",
  "topic_domain": "law",
  "word_count": 847,
  "char_count": 5231
}

Batch annotate

Annotate up to 50 documents in a single request.

Deontic classification only

Lightweight endpoint: returns only deontic modality classification.

{
  "language": "en",
  "annotation": {
    "modality": "prohibition",
    "confidence": 0.92
  }
}

Deontic modalities

Service metadata

List all registered data source annotators and their product codes.

Annotation counts from R2 sidecar metadata, grouped by product.

Full provenance audit: total annotated documents, per-product breakdown, provenance tier distribution (NATIVE 1.0, PARSED 0.95, KEYWORD ≤0.9).

Indexing API LIVE

Hybrid semantic + BM25 search, alert monitoring, and document health checks across all 20 data sources. Split across two service instances (A–E and E–W).

Health check

Binding smoke test. Returns status of R2, D1, Vectorize, and KV bindings for the worker’s product set.

{
  "service": "index",
  "status": "healthy",
  "bindings": {
    "R2_COMMISSION": "ok",
    "D1_COMMISSION": "ok",
    "RECIPE_ALERTS": "bound",
    "CF_AI_TOKEN": "set"
  },
  "timestamp": "2026-03-03T09:00:00Z"
}

Alerts

Query stored regulatory alerts from KV. Filter by recipe name and severity level.

{
  "count": 5,
  "filters": { "recipe": "*", "severity": "*" },
  "alerts": [...]
}

Hybrid query

Hybrid semantic (70% BGE-M3) + BM25 keyword (30%) search within a single product index. Includes DSA Article 27 ranking transparency metadata.

Backfill ADMIN

Admin-only endpoint. Index unprocessed R2 sidecars for a product. Useful after initial data seeding or to recover from indexing gaps. Requires infrastructure-level access (not available via public API keys).

{
  "product": "COMMISSION",
  "limit": 2000,
  "prefix": "en/",
  "indexed": 500,
  "errors": 3
}

Cross-language siblings

Find all language versions of a EUR-Lex document. Returns available languages and document paths.

Ranking methodology

DSA Article 27 ranking transparency. Returns algorithm weights (0.7 semantic, 0.3 keyword), manipulation resistance details, and update frequency. Cached for 24 hours.

Statistics

Document counts per product. The secondary index instance includes IATE term counts.

IATE deontic distribution

Distribution of deontic modalities across IATE terminology entries.

IATE cross-lingual translation

Look up a term in one language and retrieve translations across all 24 EU languages via IATE concept IDs.

{
  "concept_id": "C12345",
  "source_term": "tietosuoja",
  "source_language": "fi",
  "languages": 24,
  "translations": {
    "en": [{ "term": "data protection", "reliability": 4 }],
    "de": [{ "term": "Datenschutz", "reliability": 4 }]
  }
}

Model CDN LIVE

Serves ONNX models from EU storage (2,342 models). Supports HTTP range requests for large files.

Model manifest with all available models and SHA-256 checksums.

Serve model files with Accept-Ranges: bytes for resumable downloads. CORS enabled for browser-native inference via ONNX Runtime Web / Transformers.js.

Model categories

Data infrastructure

20 EU institutional data sources are ingested into per-product R2 buckets, annotated via queue-triggered workers, and indexed in Vectorize for semantic search. Each source has matching R2, Queue, D1, and Vectorize resources.

Data sources (20)

Pipeline

Documents flow through: Ingestion → EU storage (with metadata) → Event notification → Queue → Annotation service (topic + deontic classification) → Sidecar JSON. Searchable via the Laine Algorithm across all 20 indexes.

National law databases (27 countries)

27 national law adapters with source database links. Connected to EUR-Lex Sector 7 (290,172 national transposition measures linking EU directives to national implementations).

EuroVoc domains (21)

All annotations use the EU Publications Office EuroVoc thesaurus for domain classification:

04 Politics            08 Education & Comms     16 Environment
08 International        10 Business              17 Industry
10 EU Institutions      11 Agriculture           20 Energy
04 Economics            12 Law                   24 Production
20 Trade                14 Geography             28 Employment
24 Finance              16 Intl Organisations    32 Information
28 Social Affairs       20 Transport

EU AI Act transparency

All workers expose an Article 52 transparency endpoint:

{
  "ai_system": true,
  "provider": "Pauhu Ltd",
  "eu_ai_act_article": 52,
  "purpose": "...",
  "risk_category": "limited",
  "jurisdiction": "EU"
}

Access control

Pauhu uses entitlement-based access control, not volume-based rate limiting. Your seat tier determines what data you can access, not how many requests you can make.

Seat tiers

Burst protection

Paying seats have no daily request caps. Burst protection prevents abuse:

• Sustained: 3 requests/second sliding window
• Peak: 50 requests/second absolute maximum

Trial tier: 3 requests/day (IP-based), plus burst protection.

Response headers

X-Pauhu-Tier: live
Retry-After: 1          (only if burst limit hit)

Trial tier also receives X-RateLimit-Limit: 50.

Guides

Support

Technical: support@pauhu.eu

Sales: sales@pauhu.eu