โšก CPU-hosted ยท offline ยท developer-first

Turn any text into structured intelligence
intent ยท summary ยท entities

Ark Summarize reads a block of text and returns clean JSON โ€” the intent behind it, a concise extractive summary, and the entities it mentions. It runs entirely on the CPU with no GPU, no model download and no third-party AI provider, so responses come back in milliseconds and your text never leaves the server.

3
JSON fields out
0
GPUs required
<50ms
typical latency
100%
on-prem & offline

One input, three signals

Most summarizers give you a blob of text. Ark Summarize gives you a structured, machine-readable object you can route, store and act on directly.

  • intent The high-level purpose of the text โ€” is it a question, a complaint, a request, praise, a transaction, an instruction?
  • summary A concise extractive summary built from the most informative sentences, length you control.
  • entities People, organizations, places, emails, URLs, money, dates, times, phone numbers and percentages.
{ "intent": "complaint", "intentConfidence": 0.82, "summary": "The payment gateway has been down since Monday and customers cannot check out.", "entities": [ { "text": "Acme Corp", "type": "organization" }, { "text": "support@acme.com", "type": "email" }, { "text": "Monday", "type": "date" } ], "keywords": ["payment", "gateway", "checkout"] }

Why Ark Summarize

Built for privacy, speed, and developer ergonomics.

๐Ÿง 
Intent detection

A cue-based classifier weighs structural and lexical signals to label each text โ€” question, request, complaint, praise, transaction, instruction and more.

โœ‚๏ธ
Extractive summary

Sentences are ranked by term frequency with a positional prior, so the summary is faithful to the source โ€” no hallucinations, no made-up facts.

๐Ÿท๏ธ
Entity extraction

Regex extractors and a capitalised-phrase heuristic surface names, orgs, places, emails, URLs, money, dates, times, phones and percentages.

๐Ÿ”’
Your text stays put

Analysis happens on your server with a self-contained engine. Nothing is sent to OpenAI, Anthropic or any external API โ€” perfect for sensitive data.

โšก
Instant on CPU

No GPU, no multi-gigabyte model to download, no cold starts. Typical responses return in well under a tenth of a second.

๐Ÿ”€
Three engines, compare live

A fast lexical engine (default), a semantic MiniLM embedding model, and an abstractive DistilBART summarizer that generates new condensed text โ€” all on CPU. Run one or several and compare side by side. Choose per request via the API model field.

๐Ÿ”‘
Personal API keys

Every account gets a unique key sent in a header name you choose. Regenerate it anytime from your profile.

How it works

From sign-up to your first JSON response in under a minute.

1
Sign up with email

Enter your email, solve a quick CAPTCHA, and confirm the one-time code.

2
Grab your API key

A unique key is waiting on your profile, with a header name you can customize.

3
Summarize โ€” UI or API

Use the web tool or POST your text to /api/summarize from your app.

What developers build with it

๐Ÿ“จ Support triage

Auto-route incoming messages by intent and surface the customer, order and contact entities.

๐Ÿ“ฐ Feed digests

Condense articles into a few sentences and tag the people and organizations involved.

๐Ÿ—‚๏ธ Document tagging

Index large corpora with keywords and entities for fast, structured search.

A REST API developers enjoy

One endpoint, a JSON body in and structured JSON out, authenticated by your personal key. No SDK required โ€” just curl or any HTTP client.

  • JSON in, JSON out โ€” intent, summary, entities
  • Control summary length with maxSentences (1โ€“10)
  • Pick the engine with model (lexical, minilm or abstractive); defaults to lexical
  • Authenticated by a header name you choose
  • Live curl generator built into the web tool
Create an account
curl -X POST https://ark-summarize.immanuel.co/api/summarize \ -H "X-Ark-Api-Key: ark_your_personal_key" \ -H "Content-Type: application/json" \ -d '{"text":"Acme Corp reported that revenue grew 24% to $5M in 2025...","maxSentences":2}' # โ†’ { # "intent": "announcement", # "summary": "Acme Corp reported that revenue grew 24% to $5M in 2025...", # "entities": [ # { "text": "Acme Corp", "type": "organization" }, # { "text": "24%", "type": "percentage" }, # { "text": "$5M", "type": "money" } # ] # }

Ready to add text intelligence to your app?

Free to start. No credit card. Your data stays on your server.

Create your free account