Chat in private, with any model.

Every chat starts in Private mode, inside an enclave that your browser verifies first. Switch a chat to ZDR or Open for Claude, GPT, Gemini, and Grok.

The start screen of the app: the title Think in private, the badge Private · Verified, and the message box, with nothing to sign into.

Three modes, one switch.

Pick how private a chat is before you send it. The models of the chat follow the mode.

    • OpenAI
    • DeepSeek
    • Moonshot AI
    • Z.ai
    • Google
    • Meta

    Private · 6 models

    By default, your chat is encrypted on your device and opened only inside a verified enclave. Not even Thinkr can read it.

    • Anthropic
    • OpenAI
    • Google
    • xAI
    • Mistral AI
    • Qwen
    • DeepSeek
    • MiniMax

    ZDR · 118 models

    Claude, GPT, Gemini, and more. The provider reads your chat to answer it and keeps nothing.

    • Google
    • OpenAI
    • xAI
    • Black Forest Labs
    • Kuaishou
    • ByteDance Seed
    • Qwen
    • MiniMax

    Open · 100 models

    Every other model, video included. The provider reads your chat and may keep it for a while.

In ZDR and Open, Thinkr's proxy can read the chat in transit.

Everything a chat can take.

Photos, documents, the web, and your own history, in the same box.

The Actions menu of the message box: Add files or photos, Choose a thinker, Web search, Voice replies, and Privacy filter.
  • Verify it yourself

    Follow each step of the verification on your device: the release signed in a transparency log, the attestation report signed by AMD or Intel, and the encryption key bound to the enclave.

  • Images and documents

    Attach an image or a document to a chat and ask about it. The picker shows which models read images.

  • Web search

    Bring current web pages into a chat. In the private lane the enclave sends the search, so the search engine gets the query, never the chat or who asked.

  • History on your device

    Chats stay in your browser. Turn on backup to keep an encrypted copy that only your passkey or recovery code opens. A temporary chat leaves no history at all.

The Verification Center of the app: the hardware attested by AMD and NVIDIA, and Runtime is isolated, Passed.

The private lane, in detail.

Every chat starts here.

Encryption
HPKE, RFC 9180
Cipher
AES-256-GCM
CPU
AMD SEV-SNP · Intel TDX
GPU
NVIDIA, confidential computing
Verification
Remote attestation
Verified by
Your browser
Failed verification
Nothing sent
Thinkr sees
Ciphertext
Prompts on disk
0

Open-weight and frontier models.

Open-weight models answer in the private lane, and frontier models with zero data retention. Each row shows its mode, its context window, and what it can do.

ModelModeContextCan
GPT-OSS 120BPrivate · TEE131KThinking
DeepSeek V4.1 FlashPrivate · TEE1MImages · Thinking
Kimi K3Private · TEE256KImages · Thinking
GLM-5.3Private · TEE1MThinking
Gemma 4 31BPrivate · TEE256KImages · Thinking
Llama 3.3 70BPrivate · TEE128KText
Claude Opus 5.5Open · ZDR1MImages · Thinking
Claude Haiku 5.5Open · ZDR1MImages · Thinking
GPT-6.1 SolOpen · ZDR1MImages · Thinking
GPT-6 LunaOpen · ZDR1MImages · Thinking
Gemini 3.8 FlashOpen · ZDR1MImages · Thinking
Grok 4.7Open · ZDR500KImages · Thinking

Make more in one app.

Get started. Think in private.

Start a private chat