Security & Data Handling
LogisticsDeck calculates warehouse and inventory metrics, uses AI to interpret the results, and keeps completed reviews available for a limited time. Here’s what travels from your browser or ChatGPT, what LogisticsDeck saves, and how connected tools get access.
How a review works
- You provide the data. In the web workspace, your browser reads the export. When you submit a preview or review, the workspace sends mapped, normalized values to LogisticsDeck. In ChatGPT, the connected tool sends the data you provide for that tool call.
- LogisticsDeck validates and calculates. The service uses programmatic calculations for metrics such as counts, rates, inventory variances, scope coverage, person-days, elapsed times, and productivity or activity rates. It identifies estimates and proxies so they can be interpreted with the right limitations.
- OpenAI interprets a limited set of evidence. LogisticsDeck sends the evidence and relevant context for that review. The review request does not give OpenAI access to your LogisticsDeck account.
- LogisticsDeck checks and returns the result. The service checks the format and evidence references before accepting findings. You can reopen a completed result while it is available in history.
What happens to your export
In the web workspace, the original CSV or Excel file stays in your browser. LogisticsDeck does not upload the original file to its Cloudflare storage or send it to the review provider. The browser can keep parsed values while you work. The server uses submitted values to validate and calculate, and the review database does not keep a separate copy of the uploaded file or its normalized rows.
Depending on your file and mappings, submitted values can include order or transfer IDs, SKU or material IDs, quantities and units, timestamps, warehouse or site fields, inventory counts, and labor hours. Operator names or IDs may be present in an export. Employee names are unnecessary for identifying activity by operator. Use an internal alias where practical, and leave out personal details and fields the review does not need.
Review diagnostics record events and error categories. The review code does not log original exports or raw submitted rows. Some validation failures can produce a redacted example of the generated response for troubleshooting. This is separate from review history. The web workspace’s file handling does not describe how ChatGPT handles files you upload directly to ChatGPT.
What OpenAI receives
LogisticsDeck uses the OpenAI API from its server to interpret calculated evidence, produce structured findings and recommendations, and answer included clarifications. The review request sends a bounded set of evidence and relevant context, which can include labels and values derived from your data and an operational note you choose to include. It does not send the original file or the raw export rows. A clarification sends your question and a limited context drawn from the completed review.
If you ask for help interpreting a custom metric in Operations Review, LogisticsDeck may send OpenAI your metric description, operation type, and the types of fields you mapped. This request excludes the original file, source rows, and cell samples. Where enabled, optional shared context discovery in Operations or Inventory may also send source field names, aggregate type/shape and cardinality summaries, date/time characteristics, observed relationship hints, and existing confirmed meanings to OpenAI. This request does not send the original file, source rows, or cell samples. Shared discovery includes operation/domain hints only when independently confirmed. Where AI-assisted Operations setup is enabled, a separate setup request can send structural metadata, up to three redacted sample records covering up to eight fields, and up to 240 characters of operational context you choose to provide. Some values, such as quantities or units, can remain visible when they help explain the data. Redaction reduces exposure. Samples may still contain operational information. The original file stays in your browser. Paid analysis uses the separate validated evidence workflow described above. The setup service can reuse a validated proposal for up to 24 hours. It does not store a copy of the transmitted sample or the setup note. AI proposes context rather than formulas or KPI values; you confirm the meanings, and deterministic logic calculates the metrics. If interpretation is unavailable or uncertain, deterministic and manual setup remain available.
LogisticsDeck keeps its API key on the server, requests structured responses, and sets store: false on setup interpretation, review, and clarification requests. OpenAI says API data is not used to train its models by default. OpenAI’s own data handling and any applicable retention are separate from LogisticsDeck’s review history. See OpenAI’s API data controls.
What LogisticsDeck keeps
- Interactive Operations workspaces, where available. LogisticsDeck validates the mapped values and makes the saved workspace available for 24 hours. The saved workspace includes calculated metrics, supported breakdowns, confirmed setup and a content fingerprint. Expired workspace content is removed by scheduled cleanup. Filters and source inspection may require the original export or a matching re-upload. Unlocking, reopening and deterministic exploration do not send a findings request to OpenAI.
- Completed results. Findings, calculated metrics, and cited evidence are available for 24 hours after the result is saved. Included clarification questions and answers share the original review’s expiry. Reopening an available result uses the saved copy.
- Review and account records. Review status, timing, provider request and usage metadata, account identifiers, and credit and billing records are stored separately from the result. They can remain after the result expires. See the Privacy Policy for account closure and retention.
- Profile-linked setup. When you run a review with a saved profile, LogisticsDeck records a setup snapshot. It can include the scope, selected measurements, and labor assumptions or totals used. This record is separate from the 24-hour result cache.
Saved Operational Profiles
Profiles save reusable setup, such as a profile name, review type, site context, objective, column headers and mappings, scope choices, selected measurements, and reusable assumptions. They do not contain the original export or its row-by-row data. Profiles are saved for reuse and can be deleted in the workspace. Review setup snapshots and profile audit records are separate.
Sign-in and ChatGPT connections
Clerk handles LogisticsDeck sign-in. A supported ChatGPT connection uses OAuth: you sign in with LogisticsDeck and approve the permissions requested by the connected client. The client uses a scoped token tied to your account. Your LogisticsDeck password is not passed to the MCP client.
Approved permissions determine whether a connected client can read account status, retrieve reviews, run analyses, or read saved profiles. LogisticsDeck stores token hashes and grant information and records tool and outcome metadata for auditing. Access tokens expire after one hour. Where authorized, refresh tokens have a 30-day lifetime. You can revoke access in Connections. LogisticsDeck’s cleanup removes MCP tool audit entries after 30 days. Revocation stops future access through that grant and does not remove data already shared with the client. See Connect LogisticsDeck to ChatGPT.
Questions about your data
Contact hello@logisticsdeck.ai with questions about what a review needs, what it sends, or what remains after it expires.