ARAGATS4K · SELF-HOSTED AI GATEWAY AND KNOWLEDGE LAYER

The right model for every question, and proof for every answer.

Aragats4K sits between your people and your AI models. For each question it picks the cheapest model that meets your quality and speed targets, answers from your own documents with citations it checks itself, and keeps every request within budgets, access rights and an audit trail.

Runs on your own infrastructure · Local open models, optional cloud model · Every answer audited

Most AI deployments trade cost, speed and trust against each other.

One model for everything

Send every question to the strongest model and answers are slow and expensive. Send them all to a small one and the hard questions come back wrong.

Answers nobody can check

Models answer confidently from memory, and a citation can point to a passage that doesn't say what the answer claims.

No control over who sees and who spends

Documents leak across teams through the assistant, and spending is known only when the invoice arrives.

Every question takes the cheapest route that is good enough.

Each step shows live in the console, so you can see why a question went where it did.

1

Understand

A small classifier labels the question: task, complexity and whether documents are needed, in about 0.4 s.

2

Search

A hybrid search, by meaning and by keyword, runs over the documents this person may read.

3

Route

It picks the cheapest model that meets the workflow's quality and speed targets, based on measured results, not price lists.

4

Answer

The model answers from the passages it was given and cites each one, like [S1].

5

Verify

The server checks every citation and every cited claim against its source. A failed answer is retried once on a stronger model, then logged.

Within two points of the strongest model, at a fifth of the cost.

Measured in October 2026 on a product-support knowledge base, with local open models. Costs for local models are notional prices set in the gateway.

98%
Answers correct, routed
On a 42-question test set. Sending every question to the strongest model scored 100%.
~1/5
Of the cost
387 against 1,846 millionths of a dollar per answer, compared with always using the strongest model.
2.0 s
Median response
Against 25.5 s when the strongest model answers every question.
25/27
Wrong answers stopped
Deliberately wrong answers caught by the claim check (25 and 26 of 27 in two runs), while 35 of 36 correct answers passed.
0/10
Wrong cache hits
Look-alike questions with different answers ("within" versus "after 30 days") were never answered from the cache. True rewordings came back in about 0.4 s.
15 min
Audit anchoring
The audit log is append-only and hash-chained, and its head is copied outside the system every 15 minutes, so edits show.

Single runs vary by about one question. We re-measure on your own documents during a pilot, and the router re-checks itself every night.

A knowledge layer your teams can trust.

Your documents, any format

PDF (scanned pages read by OCR), Word, Excel, CSV, Markdown, text and images. Answers cite the page, and every document keeps its versions.

Access that follows your teams

Groups, repositories with read, write and manage rights, and public, group or private documents. Private means private, administrators included.

Knowledge base first

Per workflow: documents only, documents first, or automatic. Answers from model memory are labelled unverified, and Wikipedia can be added with citations.

Answer cache that checks first

Repeated and reworded questions are answered from the cache only when a model confirms both have the same answer. Only verified answers are kept.

Budgets before the bill

Each workflow has a spending limit in the gateway, and each person and API key a monthly one. Over-budget requests are refused before they reach a model.

A dashboard for the whole picture

Cost and the cost avoided by routing, cache hit rate, the share answered from documents, where the time goes, and the questions your documents can't answer yet.

Security built in

Single sign-on with roles, hashed API keys, credentials stripped from logs, personal data masked, and an audit log that can't be edited quietly.

Your models, your hardware

Local open models run on your own GPUs through one gateway; a cloud model can be added as an extra tier. Apps connect through an API and a command-line tool.

For teams that answer questions from documents.

Support and knowledge teams

Answers from manuals, policies and tickets, with the source on every answer and a list of what the documents don't cover yet.

IT and platform teams

One gateway for every model and app, with budgets per workflow and per person, and evidence for which model to use where.

Regulated organisations

Self-hosted, with access control per document and an audit trail that shows who asked what, which sources were used and what it cost.

Cheaper answers you can check, on infrastructure you control.