Connect an AI assistant to KlaritiQ

Ask your own AI assistant about your organisation's assurance record, and get answers that point back at the evidence behind them. An assistant such as Claude, ChatGPT or Copilot connects to KlaritiQ as the person who signs in, reads only what that person can already see, and cannot change anything.

This is different from Connect coding agents with MCP. That page is for a coding agent with a pipeline key bound to one AI system. This page is for people asking questions about the record.

What an assistant can and cannot do

An assistant connected here can read:

• initiatives, with their status and target dates,
• risks, with severity and the evidence behind each,
• evidence, as provenance only: the title, type, trust level and who asserted it. It never gets a document's contents,
• decisions and the evidence each rested on,
• AI system versions: whether a system still matches what was signed off, and exactly what changed, part by part,
• the gaps KlaritiQ's own rules found, the statutory deadline clocks, what is waiting on the person, and what changed since a date.

It cannot change anything. It cannot sign anything off, record evidence, complete a task or resolve a risk. Every tool it has is read only, and a test fails if any of them could write.

How an owner turns it on

Assistants are off until an owner turns them on.

1. Open Integrations and find the AI assistants card.
2. Choose Turn on and read the confirmation. It says plainly that what an assistant reads is sent to that assistant's provider, not to a KlaritiQ sub-processor.
3. Copy the address the card shows and give it to the people who will use it.

An owner can turn it off at any time. Every connected assistant stops on its next request.

How a person connects one

1. In your assistant, add a custom connector or MCP server and paste the address from the card.
2. The assistant sends you to KlaritiQ to sign in, as you always do.
3. KlaritiQ shows who is asking and what it can do. Choose Allow.

You need an owner, admin or member seat. A read-only seat cannot connect an assistant.

What it sees depends on who you are

An assistant acts as you. If your organisation limits you to one department, the assistant sees only that department's records plus the organisation-wide ones, exactly as you would on the website. It never sees more than you do.

Every request is checked again, at the moment it is made, against your membership and the organisation's switch. Removing you from the organisation, or turning assistants off, stops the assistant on its next request.

Seeing and stopping assistants

Owners and admins can see, on the AI assistants card:

• who has connected an assistant, which assistant, when it was last used and how many requests it has made,
• recent activity: who asked, which tool, and how it ended.

The record keeps who asked, which tool, how many rows came back and how it ended. It never keeps the answer an assistant was shown, and for a search it keeps only the length of what was typed.

An owner can Stop one person's one assistant in this organisation. It takes effect on the next request, and Allow again reverses it. This does not sign the person out of KlaritiQ or affect any other organisation they belong to.

How to read an assistant's answers

• Every answer carries the ids and links of the records it rests on. Ask the assistant to cite them, and open them in KlaritiQ.
• Titles, claims and reasons were written by people or taken from uploaded documents. The assistant is told to treat them as data, never as instructions.
• AI system drift compares recorded versions only. A change made to a live system that nobody and no pipeline recorded as a new version is not visible to KlaritiQ, so not to the assistant either.
• If a part of an answer could not be read, for example the DPDP clocks, the answer says so. "No deadlines" is only ever said when everything was read.

Limits

• Each person can make 600 requests an hour through assistants.
• A request that comes from a web page is refused. Assistants connect from their own servers.

What data leaves KlaritiQ

Records an assistant reads are sent to the assistant's provider (for example Anthropic, OpenAI or Microsoft), under that provider's terms with you. KlaritiQ does not send anything on its own: the assistant asks, and KlaritiQ answers as the person. Some regulated organisations restrict sending compliance data to outside AI tools, which is why this is off until an owner chooses it.

For what each connection can access across KlaritiQ, see How KlaritiQ Integrations Work.

Was this article helpful?

Community Questions (0)

Still have questions?