Description
Privatemode AI is built by Edgeless Systems, a German company, around one core claim: prompts and data stay encrypted through the entire AI request, including while a model is actually processing them, not only while the data is moving over the network or sitting in storage. It calls this confidential computing, and positions the service as a middle ground between using a mainstream AI provider, where prompts are processed in plaintext on that provider's servers, and self-hosting a model, which gives full control but requires dedicated hardware and ongoing operations work.
Read More
The product covers four areas: an Inference API for building AI features into another application, a web-based Chat for direct use, a coding agent installer, and speech-to-text. Requests run against a catalog of open models, including Kimi-K2.6, GLM-5.3 and a faster GLM-5.3-Flash variant, and gpt-oss-120b for chat, plus DeepSeek-OCR-2 for document OCR, Qwen3-Embedding for embeddings, and Voxtral Mini and Whisper large-v3 for speech-to-text.
Privatemode's privacy claims rest on three things it says can be checked rather than taken on trust: end-to-end confidentiality through runtime encryption, end-to-end verifiability through hardware-based certificates that prove the integrity of its backend, and external confirmation through independent audits. The company says it has passed two independent security evaluations, one penetration test and one architectural review, and that its build process is open source and reproducible, with transparency logs and remote attestation so a customer can verify the backend independently. It also reports meeting Germany's BSI C5:2026 criteria at the "very strong attestation" level for confidential computing in the cloud, and says a review recommended it for professionals bound by German confidentiality law (StGB paragraph 203), such as doctors and lawyers. The service is ISO 27001 certified, GDPR compliant, and hosted in the EU.
Pricing starts with a free tier: no credit card required, 5 million chat tokens on sign-up plus more tokens monthly, with access to Chat and the Inference API under the same end-to-end encryption. A pay-as-you-go tier bills usage per token, starting from roughly €1.27 per million output tokens depending on the model, with higher rate limits. An Enterprise tier adds custom SLAs, custom models, single sign-on, and integration support, priced on request. Privatemode also lists integrations with n8n, Anything LLM, and PrivateGPT, and publishes case studies with organizations including Capgemini and Germany's federal employment agency.
- CategoryAI Privacy Tools
- Pricing ModelFreemium
- DeploymentCloud
- PlatformsWeb
Privatemode AI's Core Features
Confidential computing that keeps data encrypted during AI processing, not only in transit or at rest
Inference API, web Chat, a coding agent, and speech-to-text in one service
Catalog of open models: Kimi-K2.6, GLM-5.3 and GLM-5.3-Flash, gpt-oss-120b, DeepSeek-OCR-2, Qwen3-Embedding, Voxtral Mini, and Whisper large-v3
Hardware-based attestation and reproducible, open-source builds for independent verification
Independent security audits, including penetration testing and architectural review
Meets Germany's BSI C5:2026 confidential-computing attestation criteria
ISO 27001 certified, GDPR compliant, hosted in the EU
Free tier with 5 million chat tokens on sign-up and no credit card required
How to use Privatemode AI?
Sign up: create a free account, no credit card required.
Try Chat: use the web-based Chat app to test a model with your own prompts.
Get an API key: switch to the Inference API to call models from your own application.
Install the coding agent: set up Privatemode's coding tool if you want encrypted AI assistance in your IDE.
Move to pay-as-you-go: upgrade once you exceed the free monthly token allowance.
Talk to sales for Enterprise: request custom SLAs, models, and SSO for larger deployments.
Privatemode AI's Use Cases
- Confidential chat and inference
- Regulated professional use
- Encrypted coding assistance
- Confidential speech-to-text and OCR
