Sovereign Drive — Tester FAQ

A local, encrypted personal knowledge vault that indexes your files (docs, PDFs, transcripts, notes) and exposes them to any AI agent you connect — Claude Desktop, Cursor, others — over MCP, so you don't re-explain your context to every tool separately.

Everything is stored in a local SQLite database in the folder you pick during setup, encrypted at rest (SQLCipher, key held in your OS credential store — not in the database file itself). Search and entity extraction run against a local embedding model by default — no API calls, no cloud.

Your files and vault content stay on your machine. A few things do reach the network: (1) the app can check whether a newer version exists, which is a small anonymous request to GitHub that sends none of your data; (2) on Windows, the speech-to-text tool is downloaded the first time you transcribe audio or video; and (3) device sync, only if you turn it on, which sends your data between your own paired devices, encrypted, directly when possible and otherwise through a relay that only passes encrypted bytes. Connected agents can read whatever you grant them, so choose which agents you connect and what they can access.

By default, only your own computer can use it. Sovereign Drive still listens on all of your computer's network addresses (0.0.0.0:3001, configurable), but unless you turn on Allow network access (Agents → Connect), a request from another device is refused with a message saying how to get access, and nothing from your vault is served to it. New installs start with network access off. A copy that predates this setting keeps network access on until you turn it off.

With network access on, other devices need the pairing code shown on the Agents page, and you can mint a token for an individual agent there. Software running on your own computer is trusted without a code, web pages open in your browser are refused outright, and there is no full sign-in system yet (a known, tracked gap). Don't turn network access on for a network you don't trust. Windows Firewall may ask about network access the first time the app runs; that prompt is about the listening address, and the setting above is what decides who can actually use it.

You can also switch the connector off entirely (the plug icon in the top bar, or the Agents page): connected agents are cut off and it stays off after a restart. The app still answers on the port while it is off, but only with a notice and no vault data. See "My agent says the connector is off" below.

This isn't a bad download — it's macOS Gatekeeper blocking the app because it isn't code-signed and notarized by Apple yet.

Fix: open Terminal and run xattr -cr "/Applications/Sovereign Drive.app", then open the app normally. No Terminal? Try opening it once (you'll hit this dialog), then go to System Settings → Privacy & Security → Security section, and click Open Anyway.

Same cause as the Mac dialog — no code-signing certificate configured yet, not a bad download. Click More info, then Run anyway.

Sovereign Drive needs the Microsoft Visual C++ runtime, which a fresh copy of Windows doesn't include. The installer (.exe) from 1.7.0 installs it for you: you'll see an administrator prompt, then a short quiet install.

Fix: if you used the .msi, or an earlier installer, install the Microsoft Visual C++ Redistributable (x64), then start Sovereign Drive again.

It means the local server that agents connect to is up. If it doesn't say RUNNING, restart the app first before troubleshooting anything else.

Fully quit and reopen the agent app after editing its config — a running session won't pick up a new MCP server without a restart. For Cursor specifically, also check it's enabled under Settings → Tools & MCP.

If it still doesn't connect, check the debug log, sovereign_mcp_debug.log, in your system temp folder: %TEMP% on Windows, and $TMPDIR (or /tmp) on macOS and Linux.

The AI tool connector was turned off on purpose (the plug icon in the top bar, or the Agents page). Sovereign Drive is running; it doesn't close the port, it keeps answering and tells the agent what happened, so you don't have to guess whether the app crashed. The agent sees a single tool, sovereign_drive_status, and every other tool call returns the same message and does nothing. No vault data is served while the connector is off.

Fix: open Sovereign Drive and turn the connector back on (the plug icon, or Agents → Connect → Start Connector), then tell your agent to try again. Agents are told not to retry on their own until you confirm.

You can also check from a browser: http://127.0.0.1:3001/health shows "status":"healthy" when the connector is on and "status":"connector_off" when it is off (use your own port if you changed it).

An agent on another device (a phone, a tablet, another PC) reached Sovereign Drive, but network access is off, which is the default for new installs. Nothing from your vault was served to it.

Fix: on the computer that holds the vault, open Agents → Connect and click Allow network access. The agent also needs the pairing code shown on that page. Agents on the same computer are not affected by this setting.

Most often Sovereign Drive isn't running at all (you quit it, it crashed, or it hasn't started), so nothing is there to answer and the agent's own app shows its generic message. To tell the cases apart, open http://127.0.0.1:3001/health in a browser:

• No response: the app isn't running, or is still starting. Open Sovereign Drive and check the plug icon.
• "status":"connector_off": the connector is switched off. See the answer above.
• "status":"healthy": the app is fine. Fully quit and reopen the agent app, then check the debug log (see "I added the config to my agent…").

No, but you do need to trust the copy-paste config on the Connect Agents page and restart your agent app once. That's the extent of the technical lift by design — if it's not, that's exactly the kind of friction this test is trying to surface, so note it rather than push through silently.

A few concrete things worth doing as a tester:

• Ask your connected agent something that can only be answered from a file you ingested.
• Create a task through the agent (not the Tasks page) and check it off later.
• Ask the agent to remember something about you — a role, a preference — and check whether vault/profile.md picked it up.
• If you use a second AI tool, ask it something that needs the same context and see if it knows.

It tracks the five things Sovereign needs to be useful to a connected agent: vault folder set, MCP server running, at least one file indexed, an agent connected, and your profile filled in. If you're stuck, that list tells you what's actually missing.

It asks for confirmation first, so you won't trigger it by accident. If you confirm: it wipes the search index, knowledge graph, timeline/daily log, your facts (approved and waiting for review), your saved applications, and the "generated document" flags on files, and resets your profile back to blank. It then immediately re-scans your vault folder from scratch to rebuild what it can from the files still on disk.

Use "Change Vault Paths" to point at a different empty folder rather than clearing the database in place, if you want a truly clean slate without touching your current data.

• Calendar sync. An earlier version synced with Google and Microsoft calendars, but it was removed. Tasks keep their own start/end times and recurrence inside the app, but nothing syncs with an outside calendar.
• The in-app "newer version available" check doesn't work yet. Check the download page for new releases.
• Agent names are self-declared, so a generic name like "mcp" can be claimed by anything calling itself that. Minting a token for an agent (Agents → Agents & Tokens) makes its name exclusive.
• Knowledge graph extraction is pattern/regex-based, not LLM-based by default — it'll miss things a human would catch, especially in spreadsheets or banking-style data.
• No automated test-your-own-changes tooling yet — if you're the engineer tester and want to poke at that specifically, that's useful feedback, not a bug report.

Real-time friction and bugs → the help email address at [email protected].

End-of-day and end-of-week structured feedback → the two Google Forms (setup log, end-of-week debrief) linked in your kickoff email.