Questions, answered.

Decision models that answer typed questions about your documents in milliseconds, say how sure they are, and keep getting better at your work.

You send a document and a question with fixed answers: yes or no, pick one, or a score. You get back the answer and how sure it is. Your code acts on the sure answers and sends the unsure ones to a person or a larger model.

Link to this answer

Teams that make the same decision thousands of times a day.

• Every alert triaged the moment it arrives. The unsure ones go to an analyst.
• Every ticket routed as it comes in. The unsure ones go to a person.
• Every engineer on your bench, checked against every skill the client asked for.

Link to this answer

A decision you make again and again about a document, with fixed answers. Four examples:

• Phishing certificates: about 6 million new ones a day (public Certificate Transparency statistics). VAKYN checks every one as it appears.
• Support tickets: each one routed to the right team as it comes in.
• Staffing: every engineer checked against every skill the client asked for, evidence attached.
• Elections: every post tagged as it is published.

Link to this answer

No. It doesn't write text. It answers questions with fixed answers, each with how sure it is. AI that answers instead of talking.

Link to this answer

Keep the LLM for the hard cases. VAKYN decides the everyday ones in milliseconds and sends only the unsure ones up, so you stop paying seconds and tokens for every case.

Link to this answer

The answer is typed by construction and comes with a probability for every option, in one pass. Nothing to parse, nothing to retry.

Link to this answer

A classifier learns one fixed set of answers. VAKYN takes any question you write (yes or no, pick one, score) about any document.

Link to this answer

Intelligence thinks. Intuition decides. Psychologists describe two ways of thinking: System 1 is fast and automatic, System 2 is slow and careful (named by Keith Stanovich and Richard West, made widely known by Daniel Kahneman in Thinking, Fast and Slow).

VAKYN is System 1 for your systems: it decides the everyday cases in milliseconds and says how sure it is. The unsure ones go to System 2, an LLM or a person. The rest of what we believe is in Lore.

Link to this answer

Yes. That's why it says how sure it is, and why the VAKYN loop exists: we find the cases it gets wrong on your work and train exactly there.

Keep math and dates in code. Ask simple, literal questions. Send the unsure answers to a person.

Link to this answer

Decisions from 997 occupations in 22 domains and 41 kinds of work activity. With the VAKYN loop, we train it exactly where it fails on your work.

Link to this answer

VAKYN, hosted by us. You send API calls to https://api.vakyn.com with an API key, and each call is paid from your organization's balance. Nothing to install and no graphics card to rent.

Link to this answer

1. Sign in with Google, GitHub or X. You get a personal organization.
2. Top up the balance from the console's billing page.
3. Create an API key and send your first call. The console's API page shows the request and the reply.

Link to this answer

Yes. VAKYN MAX speaks the Jev-compatible API, so the official TypeSafe SDKs work after you change the base URL to https://api.vakyn.com and use a VAKYN API key. VAKYN is independent and not related to TypeSafe.

Link to this answer

Yes, through organizations. Invite people to your organization from the console; they sign in with their own Google, GitHub or X account. API keys and the balance belong to the organization, and its owner manages members, keys and top-ups.

Link to this answer

We may apply rate limits to keep the service available for everyone. If you need a steady high volume, tell us at vali@vakyn.com. On your own servers there is no limit beyond your hardware.

Link to this answer

$0.0294 per million input tokens. Input tokens are the document and the questions you send. Output is free.

For example, one million calls on a 1,000-token document cost $29.40.

Link to this answer

Each organization has a prepaid balance in US dollars. Top it up with a pack of $10, $50 or $200 through Stripe. Stripe adds tax where it applies and emails the receipt. Each call is charged from the balance as it is made.

We never see or store your full card number.

Link to this answer

Yes. Your organization's first top-up is matched 100%, up to $100, as promotional credit. Top up $50 and your balance shows $100. There is no free balance before the first top-up.

Link to this answer

Redeem it on the console's billing page. The code adds its dollar amount to your organization's balance as promotional credit. Each code works once per organization, and a code can have an expiry date and a total number of uses.

Link to this answer

Promotional credit is the first top-up match, codes, and credit we grant. It is spent before the balance you paid for. It has no cash value, cannot be transferred or refunded, and may expire on a date shown when you receive it.

The balance you paid for does not expire while your account is open.

Link to this answer

Calls are refused until you top up. Your balance never goes below zero, so you are never billed for an overdraft. When the balance drops below $1, we send a low-balance email.

Link to this answer

No. No monthly fee, no minimum spend, no plan to choose. You pay for input tokens from your balance.

Link to this answer

Top-ups are not refundable, except where the law requires it or if we close your account without cause, in which case we refund the unused balance you paid for. The Terms of Service have the details.

Link to this answer

No. VAKYN MAX processes each document and question to produce the answer, then discards them. We keep neither inputs nor answers after the call returns. The one exception: when a call fails, it can sit in an error log for at most 14 days while we find out why.

Link to this answer

No. We do not use your inputs or answers to train models unless your organization agrees to it in writing.

Link to this answer

For each call: the time, the key and organization, the number of input tokens, the charge, a request id and whether it succeeded. That is how we bill you and show your usage. The Privacy Policy lists everything we collect, who handles it and for how long.

Link to this answer

Run VAKYN on your own servers with open-server: your country, your rules, and nothing is sent to us.

Link to this answer

Yes. The model weights are Apache-2.0 and the server, open-server, is MIT. The license of each model file is on its card on Hugging Face.

Link to this answer

One graphics card for the fastest results; the smallest model is built for cheap hardware. VAKYN v1 comes in three sizes: 0.8B, 2B and 4B. Install steps are in the open-server README.

Link to this answer

VAKYN-4B on open-server, Q8_0, one RTX 5090:

• 16.6 ms median for one question on a short ticket (179 tokens).
• 29.4 ms median for three questions on the same ticket.
• 1.8 s median for eight questions on a 30,232-token document.
• About 215,000 documents an hour with 10 parallel clients, 0 errors.

Link to this answer

No. It scales as far as your hardware does.

Link to this answer

The open models are the free start. You pay for VAKYN MAX if you don't want to run anything, or for the VAKYN loop: VAKYN trained on your work, on demand. For the loop, talk to us.

Link to this answer