Knowledge search
Staff ask questions in plain language and get answers from your policies, wikis and past tickets, with sources.
“What is our refund rule for annual plans?”
AI inside the tools you already run.
We add language models to the software and data you already have: search that understands questions, documents that read themselves, and assistants that act inside your tools.
Under the hood
Your data in, a sourced answer out.
Most useful AI features follow the same shape: find the right information first, then let the model reason over it.
01
Your data
Docs, tickets, PDFs, database rows
02
Indexed
Split, embedded and stored securely
03
Retrieved
The few passages that matter
04
Model
Answers using only that context
05
Your app
Answer, citation and an action
Where it earns its place
Six jobs models do well.
Document extraction
Invoices, contracts and forms are read, checked and turned into structured data in your system.
PDF invoice → line items in your accounting tool
Reply drafting
Support and sales replies are drafted from the customer's history and your guidelines, ready for a person to send.
Ticket in → draft reply in under a minute
Classification & routing
Incoming messages, leads and requests are tagged by topic, urgency and intent, then sent to the right place.
Inbox → billing, support or sales queue
Summaries
Long threads, calls and reports become short summaries with decisions and next steps pulled out.
40-message thread → five-line handover
Agents with tools
Assistants that look things up, update records and trigger workflows, within limits you set.
“Reschedule my Thursday booking”
Guardrails
What we put around the model.
A model on its own is a demo. These are what make it safe to put in front of customers and staff.
Model-agnostic
OpenAIAnthropic ClaudeGoogle GeminiOpen-source modelsYour own hosting
- 01
Grounded answers
The model answers from your approved content and cites it. When it does not know, it says so.
- 02
Evaluation before launch
We test against real questions from your business and measure accuracy, not just demo it.
- 03
Private data stays private
Personal data can be redacted before it reaches a model, and we use providers that do not train on your data.
- 04
A person in the loop
Drafts, refunds and anything sensitive wait for human approval until you choose otherwise.
- 05
Cost and rate limits
Usage is metered and capped, so a busy day does not become a surprise invoice.
Questions
Before we integrate.
The questions technical teams ask first.
Which model do you use?
Whichever fits the job. We compare models on your own examples for accuracy, speed and cost, and design the integration so the model can be swapped later.
Will it make things up?
We reduce that risk by grounding answers in your content, requiring citations, and testing before launch. For anything consequential, a person reviews the output first.
Can it work with our existing software?
Yes. Most integrations sit behind an API in your current product or connect to tools like your CRM, helpdesk and document storage. No rebuild needed.
What does it cost to run?
It depends on volume and model choice. We estimate running costs during scoping and set limits so spend stays predictable.
ontix — scoping
$ ontix scope --use-case "yours"
→ reviewing your data sources…
→ picking the right model for the job…
→ estimating accuracy and running cost…
Ready when you are.