Product / UX / UI & API

Privatemode

Confidential-computing AI inference service from Edgeless Systems. Prompts are encrypted on the user's device and processed inside hardware-based Trusted Execution Environments (confidential VMs + Nvidia H100 GPUs) so data stays encrypted even during inference and the provider cannot see it, with end-to-end remote attestation making the privacy guarantee verifiable in hardware. It serves a rotating menu of open-weight models behind a chat UI and an OpenAI/Anthropic-compatible API, and is positioned for regulated industries (finance, law, healthcare).

Launched Feb 2025. The privatemode-public repo holds the auditable Trusted Computing Base under a custom source-available license (view/compile for audit only); the privatemode-proxy, web app, and internal/oss subdirectories are dual-licensed MIT. The operational service itself is proprietary/hosted. Serves third-party open-weight models (e.g. gpt-oss-120b, Gemma, Kimi K2, Qwen embeddings, Whisper/Voxtral) - trains none of its own.

Openness

2 high confidence

2.0

license: custom-source-available(audit-only)+MIT-subdirs(proxy/web/internal-oss)
source: TCB-public(github.com/edgelesssys/privatemode-public)
service: proprietary-hosted
models: third-party-open-weight

Not open source: the security-critical Trusted Computing Base is publicly readable (and the client proxy + web app are MIT), but the main license permits only viewing/compiling for audit, and the operational service is a proprietary hosted offering. More open than a black-box API, less than open-core.

https://github.com/edgelesssys/privatemode-public recorded 2026-06-17
TCB source (attestation agent, proxy, web app, SDK) under a custom audit-only license with MIT-licensed subdirectories

Adoption

2 low confidence

2.0

Announced Feb 18 2025; offers a free tier, pay-as-you-go, and enterprise plans, and is co-marketed with Capgemini for regulated industries, but no customer or usage numbers are published. New and unquantified.

https://www.edgeless.systems/blog/what-is-privatemode recorded 2026-06-17
launch date (Feb 18 2025); Edgeless Systems as maker; confidential-inference positioning

Capability

4 medium confidence

4.0

Broad model menu plus a drop-in OpenAI-compatible API, chat UI, and strong integration story. Not top-tier because it is a confidential inference relay for others' models rather than a frontier model or full platform.

https://www.privatemode.ai recorded 2026-06-17
model list, chat UI + API, integrations, confidential-compute guarantees

Unchanged since 2026-06-17 (last edited, not re-checked)