Privacy and data handling
Updated 3 October 2026. This describes the operator-managed dAtlas application, previously Orion, during recovery. It is intended for its operator and explicitly authorized accounts, not public registration.
Information accessed and used
dAtlas receives messages, chat identifiers and message metadata through Telegram. It stores conversation history, citations, reminder requests and delivery records so that context and follow-through survive process restarts.
With account consent, Google connections can access calendar lists, events and associated details, and Gmail messages, headers, bodies, labels and account metadata within the granted permissions. Existing Orion permissions may include email modification or calendar write access; permission alone is not an instruction to perform those actions. The current recovery uses bounded reads and preserves historical records. It does not automatically replay old pending actions.
Selected notes, calendar and reminder projections, documents and other explicitly connected personal records may also be stored. Records include source identity, content revisions and local observation times. An observation time does not establish that the source is current.
Storage and access
Application records and source copies are held on operator-controlled computers in local databases, files, indexes and recovery backups. OAuth credentials are held on the operator’s computers, never on this public website. Local records and historical copies are not all encrypted at rest; access depends on the operator’s device and account controls. The operator manages maintenance and recovery. Current AI coding-agent procedures use scoped metadata, synthetic fixtures and locally executed readers with sanitized receipts; Google-derived source bodies are not to be sent to those agents. Record-specific access or a new provider transfer requires separate review and the operator’s affirmative authorization. Historical Orion agent/model transfers have not been fully reconciled; this notice does not certify their past handling. This website contains public information only, not personal source records or credentials.
External services and models
Telegram carries the conversation and outgoing reminders. Google processes account authorization and API requests. The answer model is Anthropic Claude, accessed through the operator’s existing authenticated Claude service. User-written questions, eligible prior context and reviewed engineering excerpts are sent to that provider. A selected historical-meeting feature additionally permits one necessary transcript excerpt, up to 3,000 characters, from either of two individually reviewed records, plus up to 2,000 characters of necessary question/context. It answers only the operator’s current private question after exact revision, transcript-range, audience and disclosure checks; it is not a whole-note, history or background upload. Generated summaries, private recording/access links and unrelated contact details are excluded. Prior assistant context requires approved citation lineage. Other historical transcripts, cached calendar/reminder projections, the project recovery assessment and unreviewed revisions remain ineligible. Relevant omissions and historical/transcription limits are disclosed. At the 3 October recovery check, the product’s existing Claude account matched the web account whose model-improvement setting was OFF; this is a current setting observation, not a historical guarantee. Provider handling and retention remain separate from local retention; turning off local CLI session persistence does not establish zero provider retention.
Live Google ingestion is currently paused for authorization recovery. The two selected meeting records are conservatively handled as Google-protected information even though their original audio acquisition is not verified. Their limited transfer serves the operator’s requested historical answer, not general-purpose model training, advertising or unrelated effects. All other Google API content and derivations, including historical copies and unknown-lineage records, remain outside the answer-model path until a separately reviewed, disclosed use is implemented. This is a recovery implementation boundary, not a claim that all model processing is forbidden by Google policy. Historical source copies remain in custody. Earlier Orion configurations included other model and indexing integrations; their enabled state and data handling are being assessed before reuse. This page does not claim that all historical copies or provider records have been erased.
Retention, disconnecting and deletion
There is currently no universal automatic deletion period. Local source copies, conversation history, reminders, revisions and backups remain until the operator performs a scoped retention or deletion action. Disconnecting a Google account or revoking access stops authorized future API access but does not remove already stored copies or backups. It also does not delete the original data at Google or Telegram.
You can revoke the application’s access through Google Account connections. For local inspection, export, disconnection or deletion, contact the operator at me@danishlukawski.com. Deletion must account for local copies, derived indexes and backups; no immediate or automatic complete-erasure guarantee is made. Authentication failures are surfaced as paused access rather than treated as fresh synchronization.
Purpose and policy boundary
The purpose of Google access is the operator’s personal organization, retrieval and explicitly requested follow-through. dAtlas’s use and transfer of information received from Google APIs will adhere to the Google API Services User Data Policy, including its Limited Use requirements. Recovery work does not authorize advertising, resale, general-purpose model training or unrelated outward actions using that data. Any new integration or materially different use requires a reviewed implementation and updated disclosure before activation.