Security & privacy

Your Data Stays Yours

AtlasChat was built for businesses that can't compromise on privacy. We don't train models on your content, we don't share your documents, and every tenant runs in an isolated environment with encryption at rest and in transit. Here's exactly how we protect what you trust us with.

No model training

We Never Train On Your Data

Your documents, conversations, and uploaded files are never used to train, fine-tune, or otherwise improve any model, ours or anyone else's. The only thing your content ever touches is your own retrieval index and the prompt that answers your own users' questions.

When you delete a document, an integration, or a bot, the corresponding embeddings and stored files are removed from the system. No quiet retention, no shadow copies for model improvement.

No Training Use
No Data Sharing
Honest Deletion

Encryption everywhere

Encrypted In Transit and At Rest

Every connection to AtlasChat is protected by TLS 1.2+ with modern cipher suites. Once your documents reach our storage layer, they live in encrypted object storage and our PostgreSQL database with disk-level encryption. Sensitive fields, such as integration credentials and API keys, get an additional layer of application-level encryption before they ever touch the database.

TLS 1.2+
Encrypted Storage
Field-level Encryption

Tenant isolation

Strict Per-Tenant Isolation

Every customer gets their own isolated database schema and a dedicated vector collection per bot. Queries are routed by tenant identity at the application layer, and our retrieval engine cannot return content from any collection you don't own. Cross-tenant data exposure isn't a permissions check we run, it's a structural property of how the platform is built.

Isolated Schemas
Dedicated Collections
Tenant-scoped Auth

Defense in depth

How We Protect Your Workspace

Security isn't one feature, it's a layered set of guarantees. Here's what's in place across the platform.

Role-Based Access

Per-tenant RBAC lets you decide exactly who can create bots, edit prompts, manage integrations, or view conversation history. Admins, builders, and viewers each get their own permission set.

JWT & MFA

Short-lived JWT access tokens with refresh-token rotation, plus optional multi-factor authentication for tenant and admin accounts. Sessions can be revoked at any time.

Rate Limiting

Per-tenant rate limiting on every public endpoint prevents abuse, scraping, and runaway cost from compromised credentials. Limits are tunable per plan.

Content Safety

Built-in safety filtering on both inbound queries and model responses, with configurable guardrails so your bot stays on-topic and within your policies.

Conversation History

Every conversation your bot has on your website is stored under your tenant so you can review what shoppers and visitors are asking. Use it to spot gaps in your content and see which pages your bot referenced.

Safe For Your Customers

The embedded widget runs in an isolated frame on your site, never reads your visitors' cookies or login state, and only sends the question they type. Shoppers get instant answers without exposing anything they didn't intend to share.

On our own GPUs

Your Prompts Never Leave Our Infrastructure

AtlasChat answers questions using open-source models we run on our own hardware. Your content, the questions your customers ask, and the answers we generate all stay inside the AtlasChat boundary. We don't hand your business off to OpenAI, Anthropic, or anyone else, which is one less third party for you to vet and one less data-sharing agreement to worry about.

Self-Hosted Models
No Third-Party LLM
One Vendor, One Contract

Small business friendly

Enterprise-Grade Security For A Small-Business Price

Most small business sites can't afford a dedicated security team, and they shouldn't have to. AtlasChat ships with the same encryption, isolation, and access controls that Fortune 500 vendors charge five figures a year for, bundled into a plan that fits a corner shop's marketing budget. Drop the widget on your site and your customers get a private, secure conversation about your business, with zero infrastructure on your end.

SMB Pricing
Zero Infra to Manage
Customer-Safe by Default

Data lifecycle

What Happens To Your Content

From the moment you upload a document to the moment you delete it, here is the full path your data takes through AtlasChat.

1
Ingestion
Documents enter via upload, integration sync, or URL fetch. Files are encrypted immediately and parked in object storage tagged with your tenant ID.
2
Indexing
Documents are chunked, embedded by a local model, and written to a vector collection that only your bot can read from. The original text is preserved so we can show source citations.
3
Query
A user asks a question. Retrieval pulls the relevant chunks from your collection only. Those chunks plus the question are sent to the model you've configured. The answer comes back with citations to the source documents.
4
Retention
Conversation history is stored so your users can resume a thread and admins can review activity. You decide retention windows and who can see what.
5
Deletion
When you delete a document, integration, bot, or tenant, the underlying files, embeddings, and database rows are removed. Closing your account purges your workspace entirely.

Have a Specific Security Question?

Whether you're vetting AtlasChat for your storefront, working through a procurement questionnaire, or just want to know exactly where a piece of data lives, our team is happy to walk through it with you.