# Privatemode AI > Privatemode is a Confidential AI service. Your data is encrypted at rest, in transit, and in memory, so it stays protected while it is being processed — enforced by confidential computing hardware, not by policy, and verifiable through remote attestation. Privatemode is built by Edgeless Systems. It offers an OpenAI-compatible API, desktop and web apps, and self-hosted deployments. This file lists the English pages of privatemode.ai. Each one has a German counterpart at the same path prefixed with `/de`. Every page is also available as Markdown at its own URL: send the header `Accept: text/markdown`. APIs, documentation, and agent skills are listed in `/.well-known/api-catalog` and `/.well-known/agent-skills/index.json`. ## When to use Privatemode Use Privatemode when the prompts, documents, audio, or code you send to an AI model are confidential or regulated, and neither the infrastructure provider nor Privatemode's operators may be able to read them. Privatemode is designed so that they cannot, and the Privatemode proxy or SDK verifies the attestation report before it sends any data. Best-fit jobs: - Add AI to a product that handles sensitive customer data through an OpenAI-compatible or Anthropic-compatible API: [Inference API](https://www.privatemode.ai/inference-api) and [OpenAI-compatible API](https://www.privatemode.ai/openai-compatible-api). - Automate casework over confidential files, such as patient records, insurance claims and underwriting, privileged legal documents, or citizen services: [healthcare](https://www.privatemode.ai/ai-for-healthcare), [insurance](https://www.privatemode.ai/ai-for-insurance), [legal](https://www.privatemode.ai/ai-for-legal), and [public sector](https://www.privatemode.ai/ai-for-public-sector). - Run coding assistants and agents on a private codebase without exposing source code to the provider: [confidential coding](https://www.privatemode.ai/confidential-coding). - Transcribe sensitive conversations: [speech to text](https://www.privatemode.ai/speech-to-text). - Give people a private chat assistant: [chat](https://www.privatemode.ai/chat). - Use AI in Excel, Word, and PowerPoint on confidential files, through Claude's Office add-ins and Desktop Proxy: [Office integration](https://www.privatemode.ai/office-integration), with the app at [Download Desktop Proxy](https://www.privatemode.ai/desktop-download). Not the best fit: data that is not sensitive, or a workload that needs a proprietary model. Privatemode serves open models, listed under [models](https://www.privatemode.ai/models). How an agent calls Privatemode: - A person creates the API key in the portal first. There is no registration for agents, as [auth.md](https://www.privatemode.ai/auth.md) explains. - Send requests to `https://proxyless-api.privatemode.ai/v1`, for example `POST /v1/chat/completions`, with the header `Authorization: Bearer `. That is the quickest start. Use the Privatemode proxy or SDK instead when the encryption has to happen on your side and be verifiable. - In JavaScript and TypeScript, the SDK is the npm package [`privatemode-ai`](https://www.npmjs.com/package/privatemode-ai) (`npm install privatemode-ai`). It verifies the deployment and encrypts prompts on your side. From other languages, use the proxy or the proxyless API with any OpenAI-compatible client. - The steps are in the [Privatemode API skill](https://www.privatemode.ai/.well-known/agent-skills/privatemode-api/SKILL.md), and the endpoints are in the [API catalog](https://www.privatemode.ai/.well-known/api-catalog). ## Core pages - [Privatemode AI: The always encrypted AI service](https://www.privatemode.ai/): Privatemode is the confidential AI service that protects your data end-to-end. Use AI without security and privacy worries. - [Privatemode pricing: free tier, then €0.20 per 1M](https://www.privatemode.ai/pricing): Start free with 5M tokens and no credit card. Chat models from €0.20 per million tokens, with SSO, custom models, and SLAs on Enterprise. ## Solutions and comparisons - [Confidential AI for Healthcare](https://www.privatemode.ai/ai-for-healthcare): Cut documentation time with frontier AI while patient data stays encrypted in memory. Built for healthtech companies, hospitals, and care providers. - [Confidential AI for Insurance](https://www.privatemode.ai/ai-for-insurance): Automate claims and underwriting in life, health, and legal protection insurance while medical records and case files stay encrypted in memory. - [Confidential AI for Legal Tech](https://www.privatemode.ai/ai-for-legal): Build AI features into your legal software with frontier models while your customers' privileged files stay encrypted in memory. For legal tech. - [Confidential AI for the Public Sector](https://www.privatemode.ai/ai-for-public-sector): Speed up citizen services and administrative casework with frontier AI while citizens' data stays encrypted end-to-end and processing stays in the EU. - [Confidential AI in Excel, Word, and PowerPoint](https://www.privatemode.ai/office-integration): Use Claude's add-ins in Excel, Word, and PowerPoint with Privatemode's open models. What you send stays encrypted in memory while it is processed. - [Confidential speech-to-text — Transcribe without exposure](https://www.privatemode.ai/speech-to-text): Real-time transcription for your most sensitive conversations. Your audio stays encrypted in memory and is not visible to any infrastructure provider. - [Download Desktop Proxy for macOS and Windows](https://www.privatemode.ai/desktop-download): Download Desktop Proxy for macOS and Windows. Install the app, add your API key, and give your Office add-ins, coding tools, and other clients a local address for Confidential AI. - [Is AI private? How Privatemode protects your data](https://www.privatemode.ai/security-and-encryption): Most AI services process your prompts in plaintext. In Privatemode they stay encrypted in memory, and you can verify that through remote attestation. - [OpenAI-compatible API for Confidential AI](https://www.privatemode.ai/openai-compatible-api): An OpenAI-compatible API for Confidential AI. Integrate leading LLMs while prompts, context, and responses stay encrypted in memory. - [Private AI API with end-to-end encryption](https://www.privatemode.ai/inference-api): Add private AI to your product with an OpenAI-compatible API. Prompts stay encrypted in memory and operator access is excluded by design. - [Private AI chatbot, always encrypted](https://www.privatemode.ai/chat): Privatemode is a private AI chat service. Your prompts stay encrypted in memory, and the chatbot runs inside confidential computing you can verify. - [Private AI coding agents for your codebase](https://www.privatemode.ai/confidential-coding): AI coding agents and assistants that work on your private codebase, with your source code encrypted in memory and hidden from the provider. - [Privatemode vs. ChatGPT and Claude](https://www.privatemode.ai/comparisons/privatemode-vs-chatgpt-claude): How Privatemode compares to ChatGPT and Claude for sensitive business data: leading models, without exposing prompts, files, or responses. - [Privatemode vs. Chinese AI providers](https://www.privatemode.ai/comparisons/privatemode-vs-chinese-ai-providers): DeepSeek, Qwen, Kimi, and GLM run under Chinese law. Privatemode self-hosts open-weight Chinese models like GLM in EU confidential computing instead. - [Privatemode vs. Langdock](https://www.privatemode.ai/comparisons/privatemode-vs-langdock): Enterprise AI gateways route to the same model providers behind a compliance layer. See how Privatemode removes plaintext access entirely. - [Privatemode vs. self-hosted AI](https://www.privatemode.ai/comparisons/privatemode-vs-self-hosted): Self-hosting AI gives you data control at the cost of hardware, MLOps, and months of setup. Privatemode delivers the same protection in minutes. - [Sovereign AI without the trade-off](https://www.privatemode.ai/sovereign-ai): Sovereignty is not only about where data is hosted, but about who can read it. Privatemode runs leading open models in the EU with your data encrypted in memory, enforced by hardware and verifiable through remote attestation. ## Case studies - [Capgemini runs AI coding assistants on Privatemode](https://www.privatemode.ai/case-studies/capgemini-case-study): Capgemini needed cloud-scale AI coding assistants for defense and insurance clients without exposing source code. How confidential computing did it. - [eperi uses Privatemode to run coding agents confidentially](https://www.privatemode.ai/case-studies/eperi): How eperi uses Privatemode to speed up AI-assisted development without exposing sensitive source code to the model or infrastructure provider. - [German Federal Employment Agency runs AI on Privatemode](https://www.privatemode.ai/case-studies/bfa-case-study): How the Bundesagentur für Arbeit deployed a scalable AI chatbot on leading cloud models while sensitive personal data stayed encrypted in memory. - [ID relies on Privatemode for confidential AI processes](https://www.privatemode.ai/case-studies/id-gmbh-co-kgaa): How ID, a German healthcare software vendor used by over 1,400 hospitals, runs AI on GDPR-relevant data with verifiable confidentiality. - [Kauz.ai builds compliant AI chatbots on Privatemode](https://www.privatemode.ai/case-studies/kauz-ai-case-study): Data privacy rules kept banks and insurers from adopting AI chatbots. How Kauz.ai used Privatemode to keep every prompt end-to-end encrypted. - [See how HÄVG keeps AI chatbots secure with Privatemode.](https://www.privatemode.ai/case-studies/havg): HÄVG is piloting Privatemode AI to bring AI-powered documentation and knowledge queries to over 16,000 GP practices, provably secure and privacy-compliant. - [VolOptimus automates documents with Privatemode](https://www.privatemode.ai/case-studies/voloptimus-case-study): VolOptimus reads shipping and customer data from invoices and packing lists with Privatemode, keeping confidential documents encrypted in memory. ## Integrations - [Privatemode AI integration with AnythingLLM](https://www.privatemode.ai/integrations/anythingllm): Use AnythingLLM with Privatemode for encrypted document chat, RAG, and AI agents. Keep your workspace data confidential with cloud-grade models. - [Privatemode AI integration with Claude Code](https://www.privatemode.ai/coding-tool/claude-code): Use Claude Code for agentic coding with Confidential AI. Your source code and prompts stay encrypted in memory, and are never used for training. - [Privatemode AI integration with Claude for Excel, Word, and PowerPoint](https://www.privatemode.ai/integrations/claude-for-office): Use Claude's add-ins in Excel, Word, and PowerPoint with Confidential AI. Your cells, text, and slides stay encrypted in memory while they are processed. - [Privatemode AI integration with Dify](https://www.privatemode.ai/integrations/dify): Build LLM applications with Dify and keep every inference request encrypted end to end. Documents, embeddings, and conversations are stored on your infrastructure. - [Privatemode AI integration with GitLab Duo](https://www.privatemode.ai/coding-tool/gitlab-duo): Back GitLab Duo Self-Hosted with a model you control. Run Duo on Privatemode so prompts and source code stay end-to-end encrypted and confidential. - [Privatemode AI integration with GPT4All](https://www.privatemode.ai/integrations/gpt4all): Connect GPT4All to Privatemode and use models you cannot host yourself, with your prompts encrypted in memory during processing. - [Privatemode AI integration with Hermes Agent](https://www.privatemode.ai/integrations/hermes-agent): Use Privatemode as the model provider for Hermes Agent, the open-source agent with memory and skills. Its prompts stay encrypted in memory during processing. - [Privatemode AI integration with Jan](https://www.privatemode.ai/integrations/jan): Connect Jan to Privatemode and run its MCP connectors on a model large enough to drive them, with every request encrypted in memory during processing. - [Privatemode AI integration with LiteLLM](https://www.privatemode.ai/integrations/litellm): Add Privatemode as a model in your LiteLLM gateway. Requests that carry sensitive data are encrypted before they leave and stay encrypted in memory during processing. - [Privatemode AI integration with Mattermost](https://www.privatemode.ai/integrations/mattermost): Add Confidential AI agents to your self-hosted Mattermost. Connect the Agents plugin to Privatemode's end-to-end encrypted API in a few steps. - [Privatemode AI integration with Meetily](https://www.privatemode.ai/integrations/meetily): Transcribe sensitive meetings with Meetily and Privatemode. Cloud-grade accuracy, end-to-end encryption, and no audio in plaintext at the provider. - [Privatemode AI integration with n8n](https://www.privatemode.ai/integrations/n8n): Add Confidential AI to your n8n workflows. Use cloud-grade models in all AI nodes while keeping your workflow data end-to-end encrypted. - [Privatemode AI integration with Nanobrowser](https://www.privatemode.ai/integrations/nanobrowser): Integrate Privatemode AI into Nanobrowser for end-to-end encrypted browser automation. Keep your browsing data confidential while using cloud AI agents. - [Privatemode AI integration with Open WebUI](https://www.privatemode.ai/integrations/open-webui): Run Open WebUI as a self-hosted chat interface for your team, backed by Privatemode's end-to-end encrypted processing. Conversations stay private. - [Privatemode AI integration with OpenClaw](https://www.privatemode.ai/integrations/openclaw): Use OpenClaw with Privatemode to run an open-source personal AI assistant across your messaging apps with every prompt and tool call encrypted end-to-end. - [Privatemode AI integration with OpenCode](https://www.privatemode.ai/coding-tool/opencode): Use OpenCode for AI-assisted coding with end-to-end encryption. Your source code and prompts stay encrypted in memory, never used for training. - [Privatemode AI integration with Pi](https://www.privatemode.ai/coding-tool/pi): Use Pi, the open-source terminal coding agent, with encrypted AI. Source code and prompts stay encrypted in memory, and never train a model. - [Privatemode AI integration with PrivateGPT](https://www.privatemode.ai/integrations/privategpt): Use Privatemode's encrypted AI API with PrivateGPT to chat with your documents without exposing sensitive files to a conventional AI provider. - [Privatemode AI integration with Tabby](https://www.privatemode.ai/coding-tool/tabby): Integrate Privatemode into Tabby for end-to-end encrypted code completion and chat. Keep your self-hosted assistant private with cloud models. - [Privatemode AI integration with VS Code](https://www.privatemode.ai/coding-tool/vscode): Use GitHub Copilot and Cline in VS Code with end-to-end encryption. Your source code stays encrypted in memory and never trains a model. - [Privatemode AI integration with Xcode](https://www.privatemode.ai/coding-tool/xcode): Integrate Privatemode AI into Xcode for end-to-end encrypted Swift Assist, code generation, and agentic coding. No extensions required. - [Privatemode AI integration with Zed](https://www.privatemode.ai/integrations/zed): Integrate Privatemode into Zed for end-to-end encrypted AI coding. Use the Agent Panel and Inline Assistant with confidentiality. ## Blog - [When 90% means 90%: calibrating one-token decisions from GLM-5.3-Flash](https://www.privatemode.ai/blog/calibrated-decisions-glm-flash): Raw one-token probabilities from an LLM are overconfident. A temperature from the number of options fixes most of it without labels, leaving GLM-5.3-Flash slightly better calibrated than Jev given the same temperature treatment, on our 28-dataset benchmark. With 100 labels, you get prediction sets and an error bound on automated answers. - [OpenAI Private Intelligence: what it promises, and what you can verify today](https://www.privatemode.ai/blog/openai-private-intelligence): OpenAI announced Private Intelligence at DevDay 2026: zero data retention now, confidential-computing inference as a preview this fall. What the announcement says, what it leaves open, and how to check any Confidential AI claim. - [Turn GLM-5.3-Flash into a Jev-like System One model](https://www.privatemode.ai/blog/system-one-from-glm-flash): Typed decisions with a probability for every option, from GLM-5.3-Flash in a single forward pass: about as accurate as Jev, faster from Europe, and protected by confidential computing. With a playground and a benchmark on 29 datasets. - [What a real operator exclusion in Kubernetes takes](https://www.privatemode.ai/blog/confidential-containers-operator-exclusion): A technical operator exclusion in Kubernetes is possible today, but only one architecture delivers it. Plus the paper from the IT security congress. - [AI services for professionals bound by confidentiality](https://www.privatemode.ai/blog/ai-services-professional-confidentiality): How to use LLMs under § 203 StGB without the infrastructure or model provider seeing the data, using confidential computing and the right contracts. - [Apple Private Cloud Compute moves to Google Cloud](https://www.privatemode.ai/blog/apple-pcc): Apple Private Cloud Compute now runs on third-party hardware. As Confidential AI becomes standard, the open question is transparent verification. - [AI Risks: BaFin and DORA on the Secure Use of LLMs](https://www.privatemode.ai/blog/bafin-ai-risks): BaFin's ICT guidance names two protection levels for data in processing. What that means for cloud LLM deployments, and how DORA Art. 9 is met. - [Why a third-party confidential-computing layer is required](https://www.privatemode.ai/blog/third-party-cc-layer): Even when your infrastructure provider offers confidential computing, gaps remain. Where they stop, and how Contrast and Privatemode close them. - [Privacy-preserving LLM inference](https://www.privatemode.ai/blog/privacy-preserving-inference): Learn how privacy-preserving AI inference protects sensitive data when using LLMs, from cryptographic approaches to redaction. - [C5:2026 makes confidential computing a standard](https://www.privatemode.ai/blog/c5-2026-confidential-computing): The BSI's C5:2026 catalogue introduces mandatory criteria for confidential computing and remote attestation. Privatemode already meets them. - [Confidential AI in clinical care](https://www.privatemode.ai/blog/confidential-ai-in-healthcare): How AI-powered documentation is reducing the administrative burden in healthcare – and why Confidential AI is the answer for privacy and scalability. - [Confidential RAG: secure AI knowledge without data leaks](https://www.privatemode.ai/blog/confidential-rag): Deploy AI knowledge management for confidential data without exposing it to third parties. A guide to secure RAG, private processing, and privacy. - [An AI browser extension on confidential computing](https://www.privatemode.ai/blog/confidential-browser-extension): AI browsers send your browsing context to the cloud. Our open-source extension keeps it encrypted in memory instead. Explore the demo and the API. - [Secure prompt caching in Privatemode](https://www.privatemode.ai/blog/secure-prompt-caching): Prompt caching speeds up responses, but storing context creates privacy risk. How confidential computing and cache salting make it verifiable. - [Is ChatGPT safe for sensitive business data?](https://www.privatemode.ai/blog/is-chatgpt-safe): OpenAI's privacy policy leaves data sharing unclear, and local setups are costly to run. How Confidential AI gives you capability and privacy both. ## Resource overviews - [Case studies: Confidential AI in production](https://www.privatemode.ai/case-studies): How teams in healthcare, the public sector, insurance, and consulting run AI on sensitive data with Privatemode, and what it changed for them. - [Private AI integrations: VS Code, n8n, Open WebUI](https://www.privatemode.ai/integrations): Connect Privatemode to VS Code, n8n, Open WebUI, and more. Your prompts and source code stay encrypted in memory, whichever tool you use. - [Private AI models on one encrypted API](https://www.privatemode.ai/models): Run gpt-oss, GLM, DeepSeek-OCR, Qwen, Whisper, and Voxtral through one API, each served inside confidential computing with your prompts encrypted in memory. - [Privatemode blog: Confidential AI in practice](https://www.privatemode.ai/blog): Articles on Confidential AI, confidential computing, and how Privatemode keeps your data encrypted in memory while it is being processed.