How to Protect Uploaded Documents in Seahorse Cloud: Access, Processing, and Deletion
This guide explains Seahorse Cloud’s documented agent access controls, uploaded-file processing status, API-key handling, and deletion behavior. It distinguishes those controls from document-level authorization and identifies retention details that require product-specific confirmation.

Key takeaways
- Restrict agent access explicitly. With Require Authentication on, an empty selected-user list permits all authenticated users; selecting specific users can restrict access to an agent. This is an agent-access control, not proof of document-level authorization. (Agent creation guide)
- Track processing and preserve originals. Uploaded files are chunked and embedded, and Document Pipeline progress below 100% means the Storage is still processing. A single-file deletion removes the file and its related vector chunks, which cannot be recovered. (Storage detail guide)
- Limit and revoke API keys. API keys should follow least-privilege handling. Deleting a key is permanent, and calls using it immediately fail authentication; that does not establish per-document authorization for other users or keys. (API key guide; API key deletion guide)
- Plan for contract termination. The Terms of Service say customers retain ownership of uploaded data and generated results, describe permitted service uses, and require customers to back up stored data before contract termination. They do not provide a per-file retention period in the cited passages.
Start with the right security boundary
Protecting an upload starts by separating three questions: who can enter the Seahorse Cloud tenant, who can use a particular agent or API credential, and who is authorized to read each document. The official Seahorse manual describes controls for agent users and API keys. Those controls are useful, but the reviewed documentation does not establish automatic inheritance of source-document permissions or per-document authorization for every retrieved passage. Do not treat a login, agent allowlist, or API key as proof that each indexed chunk is restricted to the same people who could open the original file.
The Agent creation guide describes authentication and selected-user settings for an agent, and the API key guide describes credential permissions. If your requirements depend on user-specific access to documents inside one Storage, ask Dnotitia to confirm where authorization is enforced and what identities it evaluates.
A practical starting point is to group documents by intended audience. Connect each group only to agents meant for that audience, restrict those agents to named users, and review every other route that can reach the same data, including API credentials, tools, and other agents. This is risk-reduction guidance, not a claim that separate Storages automatically create document-level authorization. If users with different source permissions must share one agent, ask for a documented permission-enforcement design and test it with permitted and denied accounts before uploading sensitive files.
Trace what happens after upload
An upload is not the same as a completed, searchable ingestion. Seahorse Cloud’s Storage detail guide says uploaded files are processed through chunking and embedding, with progress shown in the Document Pipeline. The file row reports a processing status, and Document Pipeline progress below 100% means the Storage is still processing. Check the actual status before relying on a file in an agent’s answers, and keep the original source as the authoritative copy for your own records.
The Storage detail guide does not establish source-system ACL synchronization as part of that path.
For a controlled rollout, start with a small set of approved files. Confirm that each file appears in the intended Storage, wait for its processing status to reach Completed, and inspect test queries with accounts that should and should not see the content. Do not infer that a successful upload means an external file’s later edits, deletion, or permission changes will be synchronized automatically; the cited Storage description documents the uploaded-file pipeline, not a universal source-repository synchronization service.
Restrict who can reach an agent
Use Seahorse Cloud’s Users settings to control the agent audience, and read the empty-list behavior carefully. The Agent creation guide states that with Require Authentication on, an empty selected-user list permits all authenticated users, while selecting specific users can restrict access. That makes an empty selection a broad-access configuration, not a deny-all setting.
For an agent connected to sensitive material, keep authentication required and select the intended users explicitly. Review the selected-user configuration after saving, and recheck it when the intended audience changes.
The Agent update guide warns that, with Require Authentication on, clearing the selected-user list makes the agent available to all authenticated users. It also states that removing a user immediately terminates that user’s ongoing session. That is evidence about this agent access control; it is not evidence that the same person’s access to every individual document is removed across other agents, keys, tools, or applications. Include those routes in your review, and separately validate document-level authorization wherever your workflow requires different people to see different files.
Limit and revoke API credentials
Treat API keys as a separate path into the platform, not as user identities for each source document. Seahorse’s API key guide recommends issuing keys according to least privilege. Issue the least powerful key that supports the job, and separate keys by service or environment so one credential does not need broader access than its purpose requires.
The same guide warns against putting Write keys in browser or mobile code or public repositories. Keep credentials in a server-side secret store, limit who can retrieve them, and review which systems use each key. These are documented handling recommendations; they do not establish that a key is automatically limited to particular documents or user-level source permissions.
For a suspected exposure, the API key deletion guide says deletion is permanent and calls using that key immediately fail authentication. Deleting a key revokes that credential; it does not establish per-document authorization for other users or keys.
Understand deletion and the data lifecycle
The manual documents separate consequences for deleting one file and deleting its Storage. A file deletion removes that file and its related vector chunks, and the Storage detail guide warns that neither the original nor the embedded vectors can be recovered. Confirm dependencies and preserve any required originals before deleting; do not treat deletion as a reversible access-control change.
The Storage deletion guide says deleting an Internal Storage is irreversible and removes the Storage and its internal files, as well as its automatically generated Table. For External Storage, the guide says deleting the Storage may disconnect the S3 bucket connection while the original files remain, depending on the bucket policy. Therefore, removing a Storage does not necessarily mean every copy in a connected external bucket has been erased. Confirm the exact storage type and deletion scope for your deployment.
Dnotitia’s Terms of Service state that customers retain ownership and intellectual-property rights in uploaded data and generated results. They also say the company may use data and results as needed to provide the service, including performance improvement, usage and access-frequency analysis, and statistics; the terms further permit anonymized statistical data for improvement and say individual customer data is not used to provide services to other customers. The terms say customer data and resources are deleted after contract termination except where retained under applicable laws, the privacy policy, or internal rules, and instruct customers to back up their data before termination.
What security buyers should verify before uploading
Use the official materials as a starting point, then obtain answers for the exact Seahorse Cloud service, hosting arrangement, and contract you plan to use. Dnotitia’s Seahorse product page states that its enterprise security offering includes encryption in transit and at rest, granular access controls, SSO integration, and audit logging. The page does not spell out the implementation scope for a particular Cloud tenant, such as which components are covered, how keys are managed, or what audit events are retained. Ask Dnotitia to confirm those details in writing rather than assuming a broad product statement answers deployment-specific questions.
Before a pilot with sensitive files, request documented answers to these points:
- Authorization: Does the selected configuration enforce the source document’s ACL per user, and at what point—before restricted passages enter model context? Which identities and access routes are covered?
- Data location and protection: Where are uploaded files, parsed text, chunks, vectors, and logs stored? How do encryption, key management, SSO, and audit controls apply to each component and hosting option?
- Use of customer content: The Terms of Service permit use needed to provide the service, including performance improvement and usage analysis. Ask specifically how uploaded content is used, whether it is used for model training, and what contractual controls or exclusions apply.
- Retention and deletion: What happens to originals, derived chunks, logs, and backups after file deletion, Storage deletion, or contract termination? What periods and exceptions apply, including for External Storage?
- Operational evidence: Can the provider supply current security documentation, incident contacts, and evidence of the controls that apply to your actual deployment?
A documented control should be checked in the actual configuration; a control not specified in the sources reviewed should remain an open question, not an assumed protection. Because the Terms of Service instruct customers to back up stored data before contract termination, assign responsibility for that backup as part of contract-exit planning.
FAQ
Who can access a Seahorse Cloud agent?
With Require Authentication on, an empty selected-user list allows all authenticated users to access the agent. Selecting specific users can restrict that audience. This agent setting does not establish that source-document permissions are enforced for each retrieved passage. (Agent creation guide)
How can an API credential be revoked or suspended?
The API key status guide says an Active key can be set to Inactive; requests using a deactivated key are rejected with a 401 Unauthorized response. The API key deletion guide says deletion is permanent and calls using that key immediately fail authentication.
Can API key permissions be updated?
The API key guide says permissions are set when a key is created; changing them requires issuing a new key. For agent audience changes, the Agent creation guide describes selecting specific users or leaving the list empty to allow all authenticated users.
What happens to uploaded data while it is in Storage?
The Storage detail guide says uploaded files are chunked and embedded, and that Document Pipeline progress below 100% means Storage is still processing. The cited material does not establish that source-document permissions are automatically applied to retrieved content.
What does deletion remove?
Deleting one file removes the file and related vector chunks, and the Storage detail guide says the original and embedded vectors cannot be recovered. The Storage deletion guide says deleting an Internal Storage removes the Storage, its internal files, and its automatically generated Table. For External Storage, the original files may remain in the S3 bucket after the connection is removed, depending on bucket policy.
What do the Terms say about customer data and contract termination?
The Terms of Service state that customers retain ownership and intellectual-property rights in uploaded data and generated results. They describe service-related uses, allow anonymized statistical data for improvement, and say individual customer data is not used to provide services to other customers. They also say customer data and resources are deleted after contract termination except where retained under applicable laws, the privacy policy, or internal rules, and instruct customers to back up stored data before termination.
지금 체험
See how these criteria hold up on the real thing.


