2kw.ai
Sign inBook a 15-min demo
AI Integration Platform

The backbone of your AI integration.

Your team is copying data from invoices by hand. Your AI costs are a black box. And switching providers means rewriting everything. 2kw.ai handles extraction, routing, and cost tracking through one API you control.

No commitment · We'll run extraction on your own documents
app.2kw.ai
eu-central-1
Operations
248K
+8.2% w/w
Spend
$1.8K
14.8M tokens
Quality
0.94
grounding
P95 latency
412ms
−6% w/w
Activity by surfacelast 7 days
MonTueWedThuFriSatSun
ExtractionsChatConversionsTranscriptions
Recent runs
BB-1047-A.dxf412ms
agent.helpdesk188ms
INV-2041.pdf298ms
report.docx→md1240ms
call_92.mp3820ms
frontier providers
6
OpenAI · Anthropic · Azure · xAI · Mistral · Ollama
input formats
20+
PDF · Office · MSG · DXF · STEP · audio · images — OCR and VLM
data residency
EU
Built-in models hosted in Europe · no cross-border processing
compliant
GDPR
All built-in models GDPR-compliant · DPA on request · audit log per request
01 · Extraction

From documents to data.

Define what data you need from a document — once. It splits large files, runs extraction across pages, and merges everything back into clean, structured output. No PDF parsing libraries. No regex nightmares.

extractions / BB-1047-A.dxf · live
FRONTSECTION A-A⌀42.5⌀28.086.0012.50H7 ±0.02AISI 304Ra 1.66× M4 ⨯ 8PART NO.REVSCALEUNITSDRAWN · mkDATE · 04·262KW.AI · TECHNICAL DRAWINGBB1m
reading geometry
extraction.resulttechnical_drawing@v2
{
"outer_diameter_mm":···,
"bore_diameter_mm":···,
"length_mm":···,
"flange_thickness":···,
"tolerance_class":···,
"material":···,
"surface_finish":···,
"thread_pattern":···,
"part_number":···,
"revision":···,
"scale":···,
"units":···,
}
grounding0.00
p50 latency0ms
20+ formatsPDFDOCXXLSXPPTXMSGEMLDXFSTEPPNGJPGTIFFMP3WAVFLAC+ more
02 · Gateway

One API, any frontier provider.

OpenAI doubles their prices. Anthropic ships something better. A client demands on-premise models. Switch providers by changing a config line — not your codebase. One OpenAI-compatible endpoint, every major provider behind it. BYOK on Scale and Enterprise.

gateway · live request flow6 providers · one endpoint
routing table · last 5m2,412 requests
RequestRouted toP50Cost
model: "openai/gpt-5.1"openai/gpt-5.1298ms$0.0024
model: "anthropic/sonnet-4.5"anthropic/sonnet-4.5612ms$0.0110
model: "mistral/large-2"mistral/large-2220ms$0.0008
BYOK · tenant=acmeazure-openai/gpt-5.1340ms$0.0000
self-hosted · ollamaollama/llama-3.3480ms$0.0000
03 · Observability

Every token, every euro, every field.

Spend, operations, quality, latency, and surface health — all in one dashboard. Set budgets, trigger alerts, and keep an audit trail for every request.

app.2kw.ai
eu-central-1

Analytics

7d30d90d
Total operations
248,412
99.4% success rate
Total spend
$1,847.22
14.8M tokens
Avg quality score
0.94
128 experiment runs
Slowest p95
412ms
218ms avg
Activity
What the platform is doing, broken out by surface
Operations over time
Daily count per surface
MonTueWedThuFriSatSun
ExtractionsChatConversionsTranscriptions
Operation mix
Share per surface
248Koperations
Extractions154,015 · 62%
Chat59,618 · 24%
Conversions22,357 · 9%
Transcriptions12,422 · 5%
Plus

The boring parts, already solved.

Prompts & schemas

Versioned like code. Pinnable by label.

@draft, @staging, @production. Diff versions, pin them, roll back in one click. Schemas sit next to prompts, same lifecycle.

Datasets & experiments

Regression testing, by default.

Pin a dataset, run versions side-by-side, lock in quality. Catch regressions before they reach production — never by accident.

Evaluators

Human and machine scoring.

Programmatic evaluators on every run. Human annotation with open and axial coding when nuance matters. Quality you can defend in a review.

Traces

Every call, one span deep.

OTel-compatible OTLP endpoint, auto-instrumented on every operation. Spans merge into your existing telemetry — no new dashboard to learn.

Proven in production

Real teams. Real workloads. Skip the plumbing.

Industrial supplyDocument extraction

Hekitec

Hundreds of supplier documents and inquiries weekly. 2kw.ai extracts part numbers, quantities, specs — straight into procurement. Hours of manual work, now in the background.

Industrial IoTLLM gateway

MKD IoT Platform

Our own IoT platform runs on 2kw.ai's gateway. Operators ask questions in plain language — gateway routes to the right model, tracks the cost, returns an answer in seconds.

Build it yourself
  • Glue together PDF parsers, OCR, and prompt chains
  • Build provider abstraction layers from scratch
  • Write your own cost tracking and quota logic
  • Maintain extraction pipelines as models change
With 2kw.ai
  • One API call — structured data back in seconds
  • Switch providers with a config change
  • Cost tracking and budget alerts built in
  • Schemas version-controlled, extractions validated
Pricing

Simple, transparent pricing.

Choose the plan that fits your workload. Scale up as you grow. No hidden fees, no surprises.

Performance

For small teams getting started.

€99/ mo
  • 3 projects
  • 5 schemas
  • 5 prompts
  • 2 team members
  • 200 extraction credits / mo
  • 1,000 conversion credits / mo
  • 500K gateway tokens / mo
Subscribe
Popular
Scale

For growing teams that need more power.

€299/ mo
  • 15 projects
  • 25 schemas
  • 25 prompts
  • 10 team members
  • 500 extraction credits / mo
  • 5,000 conversion credits / mo
  • 2.5M gateway tokens / mo
  • Bring your own keys (BYOK)
  • Priority support
Subscribe
Enterprise

For organizations with advanced needs.

€499/ mo
  • Unlimited projects
  • Unlimited schemas
  • Unlimited prompts
  • 50 team members
  • 1,000 extraction credits / mo
  • 10,000 conversion credits / mo
  • 25M gateway tokens / mo
  • Bring your own keys (BYOK)
  • Priority support
Subscribe
Custom

Tailored for organizations with specific requirements, compliance needs, or high-volume workloads.

Let's talk
  • Everything in Enterprise
  • Custom limits
  • Dedicated infrastructure
  • On-premise deployment option
  • Custom integrations & onboarding
Talk to us

Frequently asked questions.

Need something not covered here? Talk to us.

What document types does 2kw.ai support?

20+ input formats: PDF, DOCX, XLSX, PPTX, ODT, CSV, XML, EPUB, MSG, EML, HTML, Markdown, TXT, AsciiDoc, and images (PNG, JPEG, TIFF, WebP, GIF, BMP, SVG). Documents convert to Markdown, plain text, JSON, or HTML. Audio transcription covers FLAC, MP3, MP4, M4A, OGG, WAV, and WebM — with output as text, JSON, SRT, or VTT subtitles. Large files split automatically, results merged into clean structured output.

How does the multi-provider gateway work?

One OpenAI-compatible endpoint that routes requests to any major AI provider — OpenAI, Azure OpenAI, Anthropic, Google Vertex AI, or self-hosted models via Ollama. Switch providers by changing a config line, not your codebase. Built-in cost tracking and budget alerts keep spending transparent.

Can I bring my own API keys (BYOK)?

Yes — on Scale and Enterprise. Connect your existing OpenAI, Anthropic, Azure, or Google credentials and route through the platform. Keys are encrypted at rest. BYOK requests don't consume your plan's token quota. Note: data residency and GDPR guarantees apply to 2kw.ai's built-in models. With BYOK, your data flows to the provider you chose under their terms.

Is there a CLI and AI-assistant integration?

Yes. All features are accessible via an MCP server that integrates 2kw.ai directly into Claude Code, Cursor, or Windsurf — your agent can update a schema, test it, and trigger extractions without leaving the editor. There's also a CLI to use 2kw.ai features and integrate them into agentic workloads.

Can I use 2kw.ai with my existing systems?

Yes. Standard REST API integrates with anything. The OpenAI-compatible interface means existing OpenAI integrations work out of the box — just change the base URL.

How does 2kw.ai ensure extraction quality?

Every extraction follows a schema you define — field names, data types, structure. Schemas are version-controlled so you can iterate without breaking production. For large documents, automatic chunking splits the input and merges results back against your schema. Async processing handles heavy workloads with retries. If something doesn't match your schema, you see it immediately.

Is there a free trial or demo?

Yes. We offer a no-commitment 15-minute demo where we run extraction on your own documents. You see real results before making any decisions.

Stop building AI infra from scratch.

Extraction, routing, and cost control out of the box — so your team can focus on what the AI actually does for your business.

Book a 15-min demo Documentation
No commitment · We'll show you extraction on your own documents