Where your data lives, who can see it, and why we can't even if we wanted to.
How it works:
Browser Extension
↓
Native Host (your Mac, Python script)
↓
Your Vault (~/projects/jarvis or ~/Documents/Obsidian)
↓
Your LLM:
- Ollama (localhost:11434)
- LM Studio (localhost:1234)
- llama.cpp (local binary)
- AWS Bedrock (your keys, direct API)
Network calls: Zero (unless you use Bedrock/OpenAI with your keys)
Our access: Zero. We never see your vault or prompts.
Can we decrypt: N/A (nothing encrypted, it's local files)
How it works:
Browser Extension
↓
Native Host (your Mac)
↓
Your Vault (local, never uploaded)
↓
Your LLM API (direct call from your machine):
- OpenAI API (your key)
- Anthropic API (your key)
- Google Gemini (your key)
- Azure OpenAI (your key)
Network calls: Only to the LLM provider YOU chose, with YOUR keys
Our access: Zero. Your machine calls OpenAI/Anthropic directly. We're not in the middle.
Can we see your keys: No. Stored in ~/.aws/credentials or ~/.config/mindgrift (local file on your Mac)
grep -r "1real.ai" jarvis-host.py → Zero hits. No calls to our servers.
How it works:
Browser Extension
↓
mindgrift API (Cloudflare Workers, edge)
↓
Your Encrypted Vault:
- Google Drive (your OAuth, encrypted before upload)
- Microsoft OneDrive (your OAuth, encrypted)
- Dropbox (your OAuth, encrypted)
- AWS S3 (your keys, encrypted)
↓
Your LLM:
- Included: 1k Haiku queries/mo (our separate Bedrock account)
- BYO: Your OpenAI/Anthropic keys (direct API from edge)
Encryption: AES-256-GCM. Key derived from your password (PBKDF2, 100k iterations)
Where key stored:
Can we decrypt: NO. Key never leaves your device.
What if we're served a warrant: We hand over encrypted blob. Without your password, it's useless.
| What | Where | Access |
|---|---|---|
| Your personal Jarvis vault | Your Mac (~/projects/jarvis) | Only you |
| Our internal tools (1REAL.ai) | Private infrastructure, separate from this product | Only us (team) |
| mindgrift public product | Nothing hosted yet. Levels 1/2 run on your machine and your own API keys | Separate, no crossover |
| Hosted vaults (Level 3) | User's Drive/OneDrive/S3 (encrypted) | User only (we can't decrypt) |
Why this matters: Your data never touches our internal systems. Public product infrastructure is isolated.
What it does: Injects sidebar, captures page context, sends to native host
What it does: Proxies encrypted vault to user's Drive/OneDrive, can't decrypt
Audit it yourself: Clone the repo, read every line, build from source. No binaries.
| What | Do we track? | Why |
|---|---|---|
| Your vault content | NO | Never uploaded (Level 1/2) or encrypted (Level 3) |
| Your prompts to LLM | NO | Direct API call from your machine (or encrypted at edge) |
| Pages you visit | NO | Extension doesn't phone home |
| Extension install | NO | Sideloaded from GitHub, not Chrome Web Store (yet) |
| Hosted vault usage (Level 3) | YES (metadata only) | Billing: # of API calls, storage used. No content. |
This website uses Google Tag Manager and Google Analytics 4 to count visits (the site owner's own traffic is excluded). The extension itself sends no telemetry and has no analytics of any kind.
Level 1/2: No impact. Your data is local.
Level 3: Attacker gets encrypted vaults. Without user passwords, useless.
Level 1/2: We have nothing to hand over.
Level 3: We hand over encrypted vault. They need user's password to decrypt.
Level 1/2: Can't access (data is local).
Level 3: Can see encrypted blob, can't decrypt without user password.
Mitigation: Minimal dependencies. Review package.json. Build from source.
Mitigation: Open source. Audit before installing. No auto-updates (yet).
Found a security issue? Email: security@1real.ai
PGP key: TODO
Bug bounty: Not yet. Public launch first, then HackerOne program.