Frequently asked questions

Short, plain answers to the questions that come up most often. Each one links to the full detail if you want to go deeper.

Getting started

Do I need a developer?

For the first setup, yes — running the discovery command and adding the widget’s script tag are quick developer tasks. After that, an admin runs everything from the dashboard with no code: the catalog, persona, knowledge, design, and team.

How long does setup take?

Most teams go from an existing app to a working assistant in an afternoon. The Quickstart walks through it end to end.

What do I need before I start?

An existing HTTP API with at least one signed-in action, an identity provider your users already log in with (Auth0, Clerk, Cognito, WorkOS, or your own OIDC), and a Syncanix workspace.

What if discovery misses an endpoint?

You can review everything it found in the dashboard and add, edit, or remove capabilities by hand. Nothing becomes available to the assistant until you approve it.

What the assistant does

What can the assistant actually do?

It answers questions from your content and takes real actions from your API — but only the capabilities you enable, and always as the signed-in user.

Can it do something my user isn't allowed to?

No. Every action runs with the user's own identity and permissions, so it can never do more than that person could do themselves. Actions that change data ask first; destructive ones are blocked unless you allow them.

Can I turn off risky actions?

Yes. Destructive actions are off by default, and you set the risk level of every capability. You can also require a re-authentication step before sensitive actions run.

What if the AI is tricked?

Its limit is the user's own permissions, so even a manipulated prompt can only attempt what that user could already do — and writes still ask for confirmation. We treat prompt manipulation as a risk we contain and test, not one we pretend is impossible.

Security & your data

Where does my data live?

In the European Union — AWS's Frankfurt region — encrypted in transit and at rest. The region is enforced at build time, so nothing deploys elsewhere.

Can another customer see my data?

No. Each workspace is isolated by several independent layers, and automated checks run nightly to prove one workspace can't reach another's data.

Do you train AI on my data?

No. For the models Syncanix runs for you, your data isn't retained or used for training, and default inference runs in the EU. You can also bring your own model key.

Can Syncanix staff read my data?

Access is least-privilege and logged — staff and systems reach only what's needed to run and support the service. We don't sell your data or use your end-users' conversations to train AI. The formal policy is in the Trust Center.

Can I get my data deleted?

Yes. You control retention windows, can clear the assistant's memory, and can request export or erasure. A Data Processing Agreement is available.

The widget

Will the widget slow down or break my site?

No. It loads a tiny launcher first and the chat only when opened, so it costs your page almost nothing up front. If the key is wrong it renders nothing rather than throwing — a broken embed can’t break your app.

Can the widget read my page or my users' data?

No. It runs sealed inside a Shadow DOM and never reads your page's content, cookies, or other scripts. It only ever works with what you explicitly hand it.

Does it match my brand?

Yes. It picks up your brand automatically and you can fine-tune colours, copy, and layout in the Widget Design Studio — no code required.

Billing & languages

What does it cost?

Plans are a monthly price plus an allowance of compute units, which meter conversation activity. See the pricing page for current tiers and allowances.

What languages are supported?

Six, out of the box: English, Spanish, French, German, Arabic, and Hebrew — with correct right-to-left layout for Arabic and Hebrew, across the widget, the chat, and this documentation.

Can my users use it from Claude or ChatGPT?

Yes. Every workspace gets its own MCP server, so users can reach the same capabilities from Claude Desktop, Cursor, ChatGPT, and other MCP clients — signed in through your identity provider.