Effective date: 2026-10-26
StratoSort Stack ("StratoSort", "the application", "we") is built and distributed by Kennedy Applied Sciences, Inc. ("we", "us"). This document explains what data StratoSort handles, where it lives, and your rights with respect to it.
If you have read nothing else, the short version is: StratoSort is local-first by design. Your files, the analyses we generate from them, and the search index we build all live on your machine. Nothing is sent to a server we operate.
StratoSort organizes files on your computer using on-device AI. To do that, the application reads the contents of files you explicitly choose, plus new files that appear in your Smart Folders while folder watching is on (the default; switch it off in Settings). It generates analyses (categories, suggested filenames, summaries), produces vector embeddings, and writes those results to a local database so search and re-organization stay fast.
We process the following categories of data:
- File contents and metadata of files you select or that land in a watched Smart Folder. Including text extracted from documents, images analyzed via vision and OCR, transcripts of audio and video files, filenames, paths, sizes, timestamps, and file format.
- AI analysis output. Suggested categories, suggested filenames, summaries, tags, and confidence scores produced by the bundled local language model (or by a cloud provider you opted into — see section 3).
- Vector embeddings of file content. Mathematical fingerprints of the documents you analyze, used by semantic search.
- Settings. Your preferences (Smart Folders, watchers, AI provider configuration, theme, model profile, etc.).
- Operational telemetry, locally only. Structured logs and crash dumps, written to disk for troubleshooting. These are never automatically uploaded.
We do not collect:
- Identity data (name, email, phone) — there is no account system.
- Payment information — the application is delivered without a billing surface.
- Browser or device fingerprinting beyond what your operating system already exposes via
process.platformand the Electron runtime.
All persistent state lives under your operating system's standard application data directory:
| OS | Application data directory |
|---|---|
| Windows | %APPDATA%\StratoSort Stack\ |
| macOS | ~/Library/Application Support/StratoSort Stack/ |
| Linux | ~/.config/StratoSort Stack/ |
Inside this directory you will find:
- The Orama vector database and full-text index (
vector-db/). - Cached AI analyses (the
analysis-history.dbSQLite database). - Your settings (
settings.jsonand related state files). - Application logs (
logs/). - Crash dumps, if any (
crash-dumps/). - Downloaded AI model files (
models/or a comparable subdirectory).
The application data directory is owned by your user account on your operating system. We have no remote copy.
StratoSort is designed to operate without network access after first run.
- First run, model download. The first time you launch StratoSort, the default text and embedding models start downloading automatically in the background unless you pick "Continue with limited AI" on the welcome screen; the vision model downloads when you choose a profile in the setup wizard. A full profile is roughly 7–10 GB, fetched from Hugging Face. Each download is fetched over HTTPS, the source host is on a hard-coded allowlist, and the file's SHA-256 hash is verified before the model is loaded. The destination of each download is visible in the install wizard.
- Speech model, on request. Audio and video transcription needs the Whisper model (about 150 MB). It is downloaded from Hugging Face only when you click Install model after a transcription reports it missing; its SHA-256 is taken from the Hugging Face repository at download time.
- Missing bundled components. If an install is missing its bundled vision runtime or OCR
language data, the application fetches them from
github.comorcdn.jsdelivr.net. See docs/NETWORK_TRANSPARENCY.md for every connection and its integrity checks. - Update checks. Installed (packaged) builds check for updates by default; development runs
never do, and
STRATOSORT_ENABLE_AUTO_UPDATES=falsedisables the updater entirely. Background checks follow theautoUpdateCheckpreference (on by default) and can be disabled in Settings → Application → Background Update Checks; a manual "Check for updates" remains available either way. When a check runs, the only network call is to the configured update feed (GitHub Releases by default), to fetch a metadata file describing the latest version. No personal data is sent. No file contents, no usage data, and no identifiers are transmitted in update checks. Microsoft Store installs never self-update — the Store manages updates. - Optional cloud AI providers. If you explicitly configure an external provider in Settings → AI
Provider (Anthropic, OpenAI, Ollama on a remote host, LM Studio, vLLM, or any other
OpenAI-compatible endpoint), then text from the specific operations you route through that
provider is sent to that provider over HTTPS. This is opt-in and off by default. Document
analysis, every chat question (which sends the matching passages, summaries and file paths,
including text read from images by OCR) and organization suggestions each follow the default
provider unless you route that one operation elsewhere in Settings → AI Provider; per-folder
overrides apply to analysis. Video analysis sends up to four frames of each video with its
transcript, but only when "Describe video keyframes" is on and you have accepted the consent text
that mentions video frames; with that setting off, no frames are extracted or sent. Embeddings —
the data structure that powers search — and the pixels of still images always stay local
regardless of provider configuration (text read from images can still be sent by chat, as above).
See
docs/AI_PROVIDERS.mdfor the full configuration surface. - Optional MCP server. If you enable the Model Context Protocol server in Settings, the
application listens on
127.0.0.1(loopback only) for local MCP clients on your machine. Connections from the public internet are not accepted. The server is bearer-token authenticated and rate-limited. Seedocs/MCP_SERVER.mdfor details.
We operate no servers, run no analytics, and integrate no third-party telemetry SDKs (no Sentry, no Mixpanel, no Segment, no Google Analytics, no fingerprinting library) in this distribution.
StratoSort writes structured application logs to the logs/ subdirectory of the application data
directory above. Logs are JSON lines with correlation IDs, used for diagnosing problems. They
include operational events such as service startup, IPC channel activity, and analysis pipeline
state — but the structured logger redacts sensitive keys (token, password, apiKey, and equivalents)
before they are written to disk.
Logs may include file paths from your filesystem (for example, the path of a file that failed to parse). They do not include the contents of the files you analyzed.
Crash dumps are written via Electron's built-in crash reporter to crash-dumps/ in the application
data directory. The crash reporter is configured with uploadToServer: false — crash dumps are
never automatically transmitted anywhere. They exist only to help you (or to be voluntarily attached
to a bug report by you).
You can open the logs folder or export a log archive at any time from Settings → Diagnostics → Troubleshooting logs (advanced experience mode). To delete logs, delete the files in that folder.
- All AI analyses are visible in the application UI on the file's detail view.
- The application data directory is a normal directory you can open in Finder / Explorer / your file manager. Nothing is hidden.
- You can export your settings via Settings → Application → "Settings backup & restore" at any time (advanced experience mode).
- Per-file: Removing a file's analysis is available in the UI; the cached analysis and its vector embedding are deleted from the local database.
- All analyses: "Clear history" in the Analysis history & statistics window deletes every cached analysis. Settings → AI Configuration → "Rebuild All Embeddings" (advanced experience mode) clears and rebuilds the vector database. Neither touches your original files.
- Everything: Quit StratoSort and delete the application data directory listed in section 2. This removes all settings, indexes, analyses, logs, crash dumps, and downloaded models.
Uninstalling the application via your operating system's installer / package manager removes the application binaries. Application data (settings, indexes, analyses, logs, models) is not automatically removed by the uninstaller — this is intentional so that reinstalling does not lose your work. To fully remove all data, also delete the application data directory listed in section 2.
Because StratoSort processes your data locally on your own device and we receive nothing, in most data-protection frameworks (including the GDPR and the CCPA) the user is the controller of their own data. We do not hold any copy of your data and have no record to provide, correct, or delete in response to a data subject request directed at us.
If you have routed any operation to an opt-in third-party AI provider (section 3), that provider processes the relevant request data under its own terms and privacy policy; we do not retain or have access to those interactions either.
You can exercise the equivalent of inspect / correct / delete rights at any time using the controls described in section 5.
StratoSort is a productivity tool for general use. It is not directed at children under 13 (United States) or under 16 (United Kingdom and the European Economic Area), and we do not knowingly collect data from children — we do not knowingly collect data from anyone, since the application operates entirely on your local machine.
See SECURITY.md for the security model, vulnerability disclosure process, and the defenses we apply to local data (Electron sandbox, ASAR integrity, IPC schema validation, path sanitization, atomic file operations with transaction journaling).
If we change this policy, the effective date at the top of the document will change and the version of the application that ships with the new policy will include the updated text. Material changes to the data we process locally — for example, if a future release introduced an opt-in remote diagnostics feature — would be flagged in the release notes and require an explicit user action to take effect.
For privacy-related questions, including data-subject requests:
- General contact: open an issue at github.com/iLevyTate/StratoSortStack/issues.
- For sensitive disclosures (security issues that touch privacy), follow the process in SECURITY.md — use GitHub Security Advisories on the repository's Security tab.
Document version: 1.3 · Effective 2026-10-26 · StratoSort Stack 5.0.0