Your files, your privacy.
For privacy questions or account requests, email info@seehash.com. Please do not send passwords or private documents. See contact details.
Browser processing
Eligible JPEG, PNG and WebP images and PDFs are processed and validated with Rust/WebAssembly on your device. Error screenshots use browser WASM OCR first. In On-device only mode, file contents are not uploaded. Automatic mode can transfer files to our edge or server when local processing cannot finish. Prepared outputs are saved in this browser for up to 24 hours, with a 100 MB / 100-file limit. Oldest results are evicted first; expired copies are removed on the next visit. Browser storage can also be cleared by you or your browser. Original uploads are not saved locally. Recent files are separated by account or guest session and are not synced to other devices. Use Recent files to delete individual copies or clear all saved copies.
Datasets, models, archives and text
The file workspace uses a separate Rust/WebAssembly engine loaded on demand. Files are read in browser workers; large Safetensors headers and Parquet footers can be inspected with partial reads. SHA-256 hashing reads complete files in small chunks. Original file contents, schema values, tensor names and query text are not sent to analytics. Model files are inspected without executing their code or unpickling Python objects. Optional Rust edge fallback is off by default; when enabled, it uploads a bounded file for a specific operation and uses the account’s cloud quota. Downloads and inspection reports use the same on-device Recent files storage limits. Saved operation recipes stay in local browser storage until you clear site data or replace the saved recipe.
Image cropping and background removal
Image edits and the optional foreground model run in your browser. No image is uploaded for cropping or background removal. The model and WebAssembly runtime are downloaded from SizeMyFile. The original stays unchanged; exported edits are saved under the same Recent files retention and storage limits. Use in compressor transfers the edited image within this browser; a temporary transfer copy is removed when the workspace consumes it, and expires after ten minutes. Subsequent compression follows your selected processing-location setting.
AI image check
The optional AI-image classifier and metadata inspection run on your device. Model files are downloaded from SizeMyFile; your image, filename and result are not uploaded or saved. Detector scores are experimental, not proof of origin. This feature does not call our cloud AI provider.
Context assistance
PDF context extraction and screenshot OCR run on your device first. If you request AI suggestions, the description or reviewed excerpt you submit is sent to our configured AI provider (DigitalOcean serverless inference in production). It is not saved as profile memory. On-device-only mode disables this transfer.
Suggested file actions
The homepage first suggests tools on your device using file types and the number of selected files. In automatic mode, when configured, TypeSafe’s Jev service can refine that ranking using file extensions, sizes and the optional description you enter. We do not send file names or file contents for this ranking. Avoid personal details in the description. On-device only disables these model requests. Our server temporarily caches recommendation results by a hash of the request for up to ten minutes; it does not store the description in that cache. You choose an action before a file is processed. Read TypeSafe’s privacy policy.
Agent connections and workflows
The optional local agent runner downloads our processing code into a fresh browser and blocks file uploads. It reads only explicitly named files in the folder you allow, then writes new result files there. Remote MCP connections upload authorized files to private storage in the United States. Inputs expire after one hour and intermediate and final results after four hours; a periodic worker removes expired copies. Remote workflows use your account’s cloud allowance even when the Rust edge performs the operation. Scoped credentials expire and can be revoked at agent connections. We store credential hashes, consent records, processing settings, checksums and usage metadata; we do not send these file contents to a language model. Your agent provider has its own data practices. Treat a download link as private until it expires.
Accounts and email
Accounts store an email address and a password hash, or a verified Google account identifier. Verification and password-reset messages go through the configured transactional email provider. Message contents are removed from our outgoing queue after delivery. Password reset revokes previous sessions.
Subscriptions and advertising
When enabled, Stripe handles subscription checkout and billing details. We store the subscription identifier and status. Free accounts and anonymous visitors may see labeled Google AdSense banner ads on marketing, pricing and utility pages, subject to Google’s approval and available inventory. Google and its advertising partners may use cookies, identifiers, IP addresses, and device and page information to deliver and measure ads. We do not supply filenames, document contents, OCR excerpts or processing results as ad targeting data. We request non-personalised ads. Pro accounts do not receive ads or load advertising scripts after their paid entitlement is confirmed.
Google’s consent platform provides Consent, Do not consent and Manage options for visitors in the EEA, UK and Switzerland. Use Ad privacy choices beside an ad placement to revisit your choice. The Google privacy message also provides applicable US opt-out choices. Where consent is required, we do not request ad units until the consent platform reports permission for Google and device storage. Declining does not stop you from using the tools. Analytics consent is separate from advertising consent.
Learn how Google uses information from sites that use its services and manage Google ad preferences in My Ad Center.
Stateless edge fallback
If browser processing cannot finish, supported files can use our stateless Rust edge service. It returns output directly to this browser and retains no file copies. The control plane keeps only job settings and validation metadata.
Server fallback
Only files that cannot be handled by the browser or edge are uploaded directly to private object storage on our hosting infrastructure. Originals are scheduled for deletion after one hour; outputs after four hours. A cleanup worker runs every minute. Failed or cancelled jobs can be deleted immediately. Previously issued download links expire after sixty seconds.
Optional usage analytics
If you allow analytics, we send page paths and tool events (processing started, completed, failed and download requested) to PostHog Cloud in the United States. Events may include processing route, compression mode and output format. A random identifier stored for your browser tab groups events; it is not linked to your account. We exclude filenames, contents, text, email addresses, query strings and error messages. No session recordings or automatic click capture are enabled. PostHog receives network requests, including your IP address; geographic enrichment is disabled.
Analytics is off until you opt in. Use Analytics preferences on any page to withdraw consent and clear the tab identifier. Browser Global Privacy Control and Do Not Track signals disable capture. Withdrawing consent stops future events; it does not delete events already sent. Your preference is stored on this device.