AI & data security
The models behind Coassemble's AI features, where they run, how your data is handled, and how the AI allowances work.
We use AI to make course creation faster, easier and more accessible, without compromising your data security. Below are answers to common questions about how we use AI and protect your information.
Which AI models does Coassemble use?
We use trusted third-party models from Google, through Google Cloud or, on our US-based servers, through OpenRouter, our AI gateway:
- Text: Gemini Flash, for document transformation, course generation, text refinement and translation.
- Images: Gemini Flash Image, for AI image generation.
- Narration: Google text-to-speech voices, with Google's Chirp voices for premium narration.
- Support chat: Ask Coassemble, our in-app support assistant, uses Gemini models through OpenRouter. It sees your chat, the help articles that match your question, and the workspace details it needs to answer.
You can review Google Gemini's terms of service and data policies here.
Where does AI processing run?
It depends on the server your workspace runs on:
- US servers: AI features use Google's Gemini models, through OpenRouter, our AI gateway.
- UK servers: Your content is stored in the UK. AI processing, including image generation and narration, runs in Google's EU region.
- Australian servers: Your content is stored in Australia, and AI text processing runs in Sydney. Image generation and narration are processed by Google outside Australia.
- EU servers: Your content is stored in the EU, and AI processing, including image generation and narration, runs in Google's EU region.
For any other server, such as a US dedicated server or self-hosting, we confirm the AI processing region on request.
How do the AI allowances work?
There are two, and they're counted in different units, which is the thing most often misread:
- Narration characters. One narration character is one character of generated narration, counted when it's generated. Standard and premium voices draw from the same pool. A course with a translation narrates each language separately, so it uses roughly twice the narration characters of the same course in one language.
- Image generation events. One event is one image generation, counted when you generate, whether or not you keep any of the pictures. Each event produces a set of options to choose from, and the number per event can change, which is why allowances are counted in events rather than pictures.
On paid plans, both allowances are yearly and soft. Going past either one doesn't stop anything: courses stay live, learners keep taking them, and you can keep building, editing and generating. On a Sandbox, allowances are one-time and stop at the limit.
Your current usage is on the Usage limits page in Settings, and the publish panel shows an estimate before you generate narration.
How does AI image generation work?
Image generation uses Gemini Flash Image and is triggered manually, so you control when images are created and when your allowance is spent. Generated images reflect your course topic rather than just the screen they sit on, and generation shows its progress as it works through a course. You'll see how many events you have left in the generation window.
Does Coassemble use my data to train AI models?
No. Your data and course content are not used to train our AI models. AI is used for inference only: neither we nor our AI sub-processors use your personal data, content, prompts or outputs to train or fine-tune any AI model. When you use a feature like transforming a document, your document is securely stored in object storage, and we keep the course context in our database so the AI can generate relevant content for you. If you build courses through the API or an embedded builder, your developers can set discardSourceDocuments so Coassemble deletes its processing copies of your source documents after the course is built. The finished course keeps the content built from them. Database backups (up to 7 days) and our providers' own short retention periods still apply. Without this option, documents stay with the course for later editing.
Who owns content created with AI?
You do. You retain full ownership of all content created in Coassemble, whether it's generated with AI or built from scratch.
How does Coassemble protect my data?
We apply the same security standards to our AI features as we do across the whole platform. Your data is encrypted in transit (TLS 1.2) and at rest (AES-256), access-controlled, and handled in line with industry best practice. You can review our security controls in the Coassemble Trust Centre.
Where is customer data stored?
By default, Coassemble runs on US infrastructure: the application is hosted in Oregon, USA, and uploaded files are stored in AWS object storage in the US (us-east-1). Region-locked servers in the US, EU, UK or Australia, dedicated servers and self-hosting are all available on the Partner plan for teams with data residency requirements. On a region-locked server your content is stored in the region you choose, and AI processing runs as described above.
Is Coassemble GDPR compliant?
Yes. Coassemble complies with GDPR requirements, though we don't currently hold a formal GDPR certification.
Does Coassemble have a SOC 2 report?
Yes. Coassemble has a SOC 2 Type 2 report covering the Security, Availability and Confidentiality criteria, and a SOC 2 Type 1 report. You can request them through the Coassemble Trust Centre.
Does Coassemble offer a Data Processing Agreement (DPA)?
Yes, on every plan. Our standard DPA is public at coassemble.com/dpa, so you don't need to ask for it. It includes terms specific to AI processing, covering inference-only use, no training or fine-tuning, and where AI requests are processed.