Privacy notice

Who is responsible

Oliya, the operator of Atlas and enteratlas.io, is the controller for account, billing, security and service-administration information. For personal information in customer workspace data processed on an organisation’s instructions, Oliya acts as a processor where applicable; the organisation remains responsible for its purposes and instructions. This privacy notice does not replace any required customer data-processing agreement. Contact support@enteratlas.io for privacy requests.

Information we use

• Account information: email address, display name, authentication identifiers, account name and Personal/Organisation setup choice.
• Companion and device information: an opaque device identity, chosen device name, registration, last-seen and revocation status.
• AI infrastructure metadata you choose to sync: artifact names and types, portable relative paths, relationships, workflow steps, counts and opaque identifiers. When locally derived descriptions are enabled, this also includes one description of up to 260 characters per file, selected locally from explicit metadata, a script synopsis or explanatory file text and retained for that file's current content hash.
• Optional AI analysis: a web request queues only workspace identifiers. The Companion picks it up during a connection check and shows the maximum batch and credit estimate before asking for consent. Approved AI reviews process redacted source regions, per-file profiles, comparison inputs and bounded evidence excerpts within the approved scope. An approved missing-description repair redacts common sensitive-value lines, sends the complete eligible file in temporary chunks, and synthesises one description from all chunk summaries. Description repair accepts files up to 2,000,000 characters and reports larger files instead of silently analysing only part of them. Atlas does not store source chunks, briefs or excerpts in its source-content database, action queue or AI usage ledger. Atlas temporarily retains structured batch results for restart recovery, never source text, and removes those checkpoints when the operation ends. Final findings and generated descriptions are retained for their current content hashes.
• Terms agreement records: your account identifier, the terms version accepted, the privacy-notice version presented, the server-recorded time and whether you agreed during registration, an invitation or sign-in. We do not collect additional IP addresses or device fingerprints for this record.
• AI usage records: feature, model, token counts, credit charge, provider cost, consent version and timestamps. These records never contain source-file content.
• Subscription information: Stripe customer and subscription identifiers, plan status, payment events and billing contact information. Atlas does not receive full card numbers.
• Security and operational information: limited request, authentication, webhook and error logs needed to protect and operate the service.
• Support correspondence sent to the Atlas support address.

Filenames, relative paths and descriptions may themselves contain confidential or personal information. Review what you connect and share. Redaction reduces exposure but cannot guarantee that every secret or sensitive detail is identified.

GitHub-hosted scanning and AI Review

When an organisation Owner connects a selected GitHub repository, Atlas uses a short-lived GitHub App installation token to check out the configured exact commit and optional folder in isolated Atlas-controlled compute. Source files exist there only while a structural scan or explicitly approved AI Review runs. For a recent-changes review, Atlas also checks out the exact commit used by the completed baseline review so it can derive the changed and directly related file scope. Atlas persists commit identifiers, structural metadata, approved scope metadata, final findings, counts and sanitized operational status—not either checkout, Git credentials, complete repository files, architectural briefs or evidence inputs. Temporary files and credentials are removed after success, failure or timeout. Atlas never writes to the repository.

Why we use it

We process account, device, source configuration, metadata and subscription information to provide the service and perform our contract with you. We use limited security and operational information for our legitimate interests in preventing abuse, diagnosing failures and protecting tenants. We retain billing records where necessary to meet legal obligations. Terms agreement records provide evidence of our contract; agreement is not blanket consent to optional AI processing, evidence upload or marketing. We do not use uploaded AI infrastructure metadata for advertising.

AI-provider processing

For an AI review, missing-description repair or image description you approve, Atlas sends the relevant bounded source text, thumbnail or instructions to Microsoft Foundry Models sold by Azure, including Azure-hosted OpenAI models, and receives structured findings or descriptions. Atlas sends requests to the configured Azure Foundry resource; Microsoft hosts and operates the model runtime within Azure and states that prompts and completions are not made available to OpenAI or other model providers. The approval screen identifies the scope and estimated Atlas-credit use before processing begins. A description repair stores the generated descriptions in Atlas and asks the registered Companion to add them only to unchanged, safely editable local files; you remain responsible for reviewing and committing those changes. Microsoft states that Foundry customer data, prompts and completions are not used to train foundation models or improve Microsoft or third-party products or services without explicit permission or instruction. Standard provider abuse-monitoring processes may retain or permit authorised review of selected prompts and responses unless the relevant Azure subscription has approved modified abuse-monitoring or equivalent controls. Atlas has not opted in to share API content for training. See Microsoft Foundry data, privacy and security.

Optional issue evidence

Short issue evidence upload is off by default for every local workspace. If its Companion operator enables it, Companion validates source locations, redacts or withholds secret-like lines, and uploads no more than 12 lines and 1,200 characters per excerpt, four excerpts per finding and 100 KB per sync. These excerpts are stored separately from metadata, Health reports, history, operational logs and AI result records. Subscribed organisation members can view them; browser clients cannot write them. Atlas keeps them only for the current source version and deletes them when replaced, when the option is disabled or when the workspace is deleted. If Companion is offline at opt-out, it retries deletion when it reconnects.

Optional image previews

Atlas Companion can detect images referenced by mapped AI ecosystem files. Image-preview upload is off by default. If you enable it, Companion creates bounded thumbnails locally and uploads only those previews and their relative paths; Atlas does not upload the original image files. A full organisation member can separately choose to send one stored thumbnail to the configured AI provider for a short description. Atlas shows the estimated credit use first, requires confirmation, stores the description only while that image remains unchanged, and sends the request with provider-side response storage disabled where the provider supports that control.

Service providers and international processing

Atlas uses Supabase for authentication and database services, Cloudflare for website delivery, isolated GitHub processing and email routing, Stripe for subscriptions and payments, GitHub for connected repository access, Resend for application emails such as organisation invitations, and Microsoft Foundry for explicitly approved AI reviews, description repairs and image descriptions. These providers process information under their own security and data-protection commitments. Information may be processed outside the UK; where required, we rely on an adequacy decision or approved contractual safeguards. Microsoft Foundry Global or Data Zone deployment types may process prompts and responses outside the Azure resource's region as described by Microsoft, while data stored at rest remains subject to Azure's applicable location and service-boundary commitments.

Retention

Account and synced metadata are kept while the account is active and for a limited period needed for recovery, security and dispute handling. Terms agreement records are kept while the account exists and deleted through the account-deletion process. Non-personal copies of each published legal-document version are retained separately. Opted-in issue evidence is retained only for the current source version. Atlas does not persist source-file contents sent temporarily for an AI review. Microsoft Foundry Responses requests use store: false, so responses are not retained as retrievable application state by that API feature; this setting does not remove Microsoft's standard content-filtering or abuse-monitoring processing described above. Atlas does not use Foundry Stored completions, Assistants threads, vector stores, fine-tuning files or Batch processing for AI reviews unless this notice is updated for that feature. Atlas temporarily keeps structured, source-free batch results so an interrupted review can resume without repeating completed AI work; these checkpoints are deleted when the review completes, fails or is cancelled, and otherwise expire after seven days. Atlas retains the final structured findings and bounded usage records. Device records remain until removed with the organization or no longer needed for security. Billing records are kept for the period required by tax, accounting and legal obligations. Support and operational records are kept only as long as reasonably needed for the issue or security purpose.

Account privacy and sharing

Personal accounts are private by default. Information shared through an organisation is available according to its membership and permissions. Service providers process information as described in this notice. Check recipients and permissions before sharing exports or inviting other people.

Your choices and rights

You can disable locally derived descriptions and short issue evidence per workspace before scanning, revoke Companion installations, cancel billing through Plan, and ask us to access, correct, erase, restrict or provide applicable personal information. You may also object where processing relies on legitimate interests and withdraw consent where consent applies. Email support@enteratlas.io. You may complain to the UK Information Commissioner’s Office.

Security and changes

Atlas uses tenant isolation, restricted service credentials, encrypted transport and Windows-protected local session storage. Synced descriptions are stored in the organisation's tenant-isolated Atlas data. We use appropriate technical and organisational safeguards, but no system can guarantee complete security. We investigate suspected incidents and notify affected people and authorities where required by law. We will update this notice when processing materially changes and show the revised date.

Approximate user geography

When an authenticated user opens Atlas, Cloudflare provides a two-letter country code derived from the network request. Atlas stores that country code with first-seen and last-seen timestamps for aggregate platform operations reporting. Atlas does not retain the associated IP address, city, region or coordinates in this feature, and the system-admin view exposes country totals rather than individual locations.