Privat3 is the AI-native IDE for macOS and Windows. The newest frontier models, zero content filters, zero telemetry — and an architecture where your code, prompts and keys never leave your control.
Enter the vaultSwitch between the latest frontier and open-weight models the week they drop. Privat3 adds no refusal layer, no hidden system prompt and no silent rewriting — your prompt reaches the model exactly as you wrote it.
Your workspace is encrypted on-device with keys only you hold. Prompts travel end-to-end encrypted, are never logged and never used for training. Telemetry isn't switched off — it was never written.
When your laptop isn't enough, inference moves to RAM-only GPU nodes running inside attested enclaves. No disks, no request logs, no backups. Close the session and the node wipes itself.
A real desktop app, not a browser tab. Metal acceleration on Apple Silicon, CUDA and DirectML on Windows. Run fully offline, or flip one switch to air-gap the entire IDE.
ghost-chat — src/crypto/ratchet.ts — Privat3No analytics SDKs, no crash reporters, no usage beacons. Point your firewall at it — Privat3 stays silent.
Cut every socket. Local models keep coding.
No injected guardrails, no lectures. Just the model.
AES-256-GCM, Argon2id-derived keys. Lose the laptop, not the code.
Stored in macOS Keychain or Windows Credential Manager. Never synced, never seen.
Run quantized open-weight models on your own hardware, then hot-swap to an encrypted enclave mid-conversation when you need more muscle.
Multi-file edits, terminal, tests. Everything runs in an on-device sandbox you can watch line by line.
Reproducible builds, an open relay protocol and a public warrant canary. Privacy you can audit, not just believe.
| Feature | Privat3 | Typical cloud AI IDE |
|---|---|---|
| Code readable by provider | Never | Yes |
| Prompts & outputs logged | Never | Often |
| Your code trains models | Never | Opt-out |
| Added content filters | None | Strict |
| Works fully offline | Yes | No |
| Telemetry | None | On by default |
