Introduction

Thinkr is an AI chat app with two lanes. In the private lane, open-weight models answer inside trusted execution environments (TEEs), and your browser verifies each enclave before it sends a word. In the open lane, frontier models such as Claude, GPT, Gemini, and Grok answer under zero data retention (ZDR), each with a label that says who can read the chat.

Every new chat starts in the private lane, and you need no account to try it.

The private laneCopy Anchor Link

A TEE keeps the memory of a workload encrypted, with keys that never leave the processor, and proves which code runs inside it. The private lane runs open-weight models, such as GPT-OSS 120B, DeepSeek V4.1 Flash, and Kimi K3, in enclaves on AMD SEV-SNP or Intel TDX, with NVIDIA GPUs in confidential computing mode.

Before each chat, your browser verifies the remote attestation of the enclave. It then encrypts your message with Hybrid Public Key Encryption (HPKE, RFC 9180) to a key that exists only inside that enclave. Thinkr's proxy forwards the ciphertext and cannot read it, so not even Thinkr can read the chats of the private lane. If the verification fails, your browser sends nothing.

The open laneCopy Anchor Link

Claude, GPT, Gemini, Grok, and other models that do not run in a TEE answer in the open lane. Each request goes through OpenRouter to a provider that promises by contract to keep no prompts and no answers. The provider reads your messages to answer them, and the label of each model says so.

Three productsCopy Anchor Link

• Chat: the app at chat.thinkr.sh, with both lanes and a free trial.
• Inference: an OpenAI-compatible API for the models of the private lane.
• Containers: your own Docker image inside a TEE.

Your chats stay with youCopy Anchor Link

You chat with no account, no email address, and no phone number. The app saves your chats in your browser, and Thinkr never receives a copy. The site and the app set no cookies and run no analytics scripts or trackers.