Security and data handling
What we do with your content, where it lives, who can reach it, and which assurances we do not yet hold. Written for the person who has to sign off on it.
The four things that matter most
Each of these is enforced in the product, not a policy statement.
Your data is yours
Content you upload or let us crawl is used for one purpose: answering questions for your own chatbots. It is never used to train an AI model, ours or anyone else's, and it is never pooled with another customer's content.
Workspaces are isolated at the database
Separation is enforced by Postgres row-level security, so a query for one workspace cannot return another workspace's rows even if application code asks it to. This is a database guarantee rather than an application check, which is the difference that matters when something goes wrong.
Connector credentials are encrypted at rest
When you connect Notion, Google Drive, Zendesk or any other source, the OAuth token is encrypted before storage. Tokens are scoped to the workspace that authorised them and can be revoked from your side at any time.
Answers are traceable
Every answer cites the page it came from, and strict mode confines the bot to content you approved. That is a security property as much as a quality one: you can always establish what the bot said and which source it used.
Where your content goes
Three hops, and nothing beyond them.
When you add a source, we fetch it, split it into passages and store those passages plus their vector embeddings against your workspace. When a visitor asks a question, we retrieve the passages that match, send them to the language model along with the question, and return the generated answer with a citation.
That means your content is transmitted to the model provider at answer time in order to produce the answer. It is not retained by us for training, and we do not share it with other customers. If you need a written description of this flow for a vendor review, ask and we will send one.
- Content at rest: stored per workspace with row-level security
- Content in transit: TLS, and sent to the model provider only as answer context
- Conversation logs: retained so you can review what the bot said, and exportable
- Deletion: removing a source removes its passages and embeddings
Who can reach your workspace
Access is scoped to people you invite, and the powerful keys never reach the browser.
Team members reach a workspace only by invitation. The privileged service-role database key is held by the backend alone and is never shipped to the browser, which receives a restricted anonymous client whose reach is bounded by the same row-level policies described above.
Platform administration is restricted to an explicit allowlist of addresses rather than a role anyone can be granted, and administrative tables carry row-level security with no public policies at all.
What we do not have yet
This section exists because you will find out anyway, and it is better that you find out here.
matram.ai holds no SOC 2 report, no ISO 27001 certificate, and no HIPAA, PCI DSS or FedRAMP attestation. We do not offer on-premise, air-gapped or private-cloud installation, dedicated single-tenant infrastructure, a contractual uptime SLA, or a named customer success manager.
If your review requires any of those, we are not a fit today, and we would rather tell you now than after a procurement cycle. Teams with those requirements typically look at the enterprise conversational-AI platforms instead. Our comparison of that tier is on the
- No SOC 2 Type I or Type II report
- No ISO 27001, HIPAA, PCI DSS or FedRAMP
- No on-premise, air-gapped or private-cloud deployment
- No dedicated single-tenant instance
- No contractual uptime SLA
GDPR and data requests
We are built to support the common data-subject requests, and we will act on them.
You control what content enters the system and can remove any source, which removes its stored passages and embeddings. Conversation history is exportable, and lead records can be exported to CSV or deleted. For a data processing agreement or a specific deletion request, email us and we will handle it directly.
We are a small team and we answer these ourselves, which in practice means a faster response than a ticket queue, and a real person who can tell you exactly what is stored.
Frequently asked questions
Questions your security review needs answered?
Email us and a real person will answer, including the awkward questions. Start with a 7-day free trial, no card required.