Sub-processors
Last updated 15 October 2026. These are the third parties that process customer data on behalf of Ontolix Pty Ltd ("Ontolix", "we") to provide the Ontolix platform. For each one we list what it does, the data it receives and where it processes that data.
Changes to this list
We notify your organisation of material additions or replacements of an infrastructure sub-processor, with a reasonable period to object, as set out in your agreement with us. Autodesk only handles a Revit model we convert for you, and your administrator accepts Autodesk's terms in Ontolix before your first conversion. This page always shows the current list.
Current sub-processors
| Sub-processor | Purpose | Data it receives | Processing location |
|---|---|---|---|
| Amazon Web Services (AWS) | Hosting, storage, databases, the project knowledge graph, sign-in, audit logging, web delivery and AI inference through Amazon Bedrock. | All customer document content and the information extracted from it, account and usage data, and sign-in identifiers. | Sydney (ap-southeast-2) for storage, compute, logs, backups and most AI inference. Two AI features use Bedrock global cross-region inference and may be processed in any supported AWS commercial region (see below). Web delivery uses the global CloudFront network. |
| Autodesk | Converting a native Revit model (.rvt) to IFC, only where you supply one. | Native Revit model files only. No drawings, specifications, extracted data or graph content. | United States. The model is read from, and the result written back to, our Sydney storage through short-lived links, and is held on Autodesk compute only for the conversion job (capped at 45 minutes). Autodesk keeps its own job status record for three days. |
| OpenRouter, routing to the Jev model served by TypeSafe | Automatic classification of each uploaded document, such as report type and discipline. | Each document's name, its folder and up to 3,000 characters of first-page text. We add no organisation, project or user identifier, but names and text you supply may contain them. | Outside Australia and outside AWS. Our policy requires the OpenRouter account to be set to zero data retention. |
| Stripe | Subscription billing and card payments, only for organisations that pay by card. Organisations billed under a contract do not use Stripe. | Your organisation's name and an internal organisation identifier, plus the billing contact, invoice and card details entered on Stripe's own payment pages. No project content. Ontolix never sees or stores card numbers. | United States and other countries where Stripe operates. |
Processing outside Australia
Everything Ontolix stores stays in Sydney, including databases, files, the knowledge graph, logs and backups. Four processing paths send customer content outside Australia:
- The default chat agent, which runs OpenAI's GPT-6 Luna on Amazon Bedrock global cross-region inference. It sees your question, the conversation so far and the project information it reads to answer, including content from connected sources such as email and Teams where your organisation has switched that on. An Ontolix operator can return your organisation to a Sydney-only model on request.
- Consultant-report reading, on the same global inference. It sees the text of the consultant reports being read.
- Document tagging through OpenRouter, which is on by default. We can switch it off for your organisation on request, and your documents then keep their automatic rule-based tags.
- Native Revit model conversion by Autodesk in the United States, only where you supply a native model. We can switch it off for your organisation on request, and extraction then runs from drawings instead.
Switching any of these off is free. Email support@ontolix.ai to ask.
About the AI models
AI inference runs on Amazon Bedrock, so your content goes to AWS and not to Anthropic's or OpenAI's own services. Bedrock does not share prompts or outputs with the model providers, and neither AWS nor Ontolix uses them to train models. AWS may keep a request that its automated abuse detection flags for up to 30 days, in the region that processed it. These models do not run with zero data retention.
Before a tenant's first Revit conversion, a tenant administrator must accept Autodesk's terms in the platform. Autodesk's published terms allow it to use content processed by its services to provide and improve its products.
Planned, not yet enabled
JetBrains YouTrack Cloud may in future hold the text of support requests and their attachments, outside Australia. It is not enabled and receives no customer data. We will give notice under this page before it is.
Services your organisation connects
These are not sub-processors. They are your organisation's own systems, and they exchange data with Ontolix only when your organisation connects them. Your organisation's agreement with each provider governs that provider's handling of your data.
- Procore, to read project documents and, where enabled, write back generated items.
- Microsoft 365, to read approved SharePoint and OneDrive files, Outlook mailboxes and Teams channels or chats.
- BuildPass, connected by each user to their own BuildPass account.
- Slack, to read the channels your organisation approves and, where allowed, post updates you approve. Posts go out as the Ontolix app, never as a person.
- Your organisation's sign-in provider, such as Microsoft Entra ID or Google Workspace, when your organisation uses single sign-on. Ontolix reads each person's account status, group membership and directory identifier to sign them in and keep their access current.
- Claude (Anthropic) and ChatGPT (OpenAI), when your administrator lets your people reach Ontolix from those assistants. The assistant receives what Ontolix returns, such as source excerpts, project lists and answers with citations. The provider decides where that is processed and stored, which may be outside Australia.
Contact
For more detail, or a copy of this list for your records, email support@ontolix.ai.