Thinkr
Try chat

Think inprivate.

A private AI platform for verifiable computation.

Example questions for the chat of Thinkr.

Find auth bugs in my repo.
Start a private chat, free

Enclave
Verified
Sealed
HPKE · AES-256-GCM
Thinkr sees
Ciphertext

An OpenAI-compatible API. The SDK checks the enclave before each call.Soon

Enclave
SEV-SNP · TDX
Measurement
Public log
Attestation
Checked by clients

Any Docker image in a hardware enclave, measured for anyone to check.Soon

  • Hardware enclaves
  • Remote attestation
  • Encryption to the enclave
Follow on X

How your chat stays private

Sealed from the first word.

AI is where you think out loud. Every chat takes this path by default, and not even Thinkr can read it.

  1. Sealed on your device

    Your device encrypts each message to a key that only the enclave holds, before anything leaves it.

    Encryption
    HPKE, RFC 9180
    Key exchange
    X25519
    Cipher
    AES-256-GCM

  2. Answered in sealed hardware

    Your device first checks the enclave's hardware proof against open-source code. If it fails, nothing is sent.

    CPU
    AMD SEV-SNP · Intel TDX
    GPU
    NVIDIA, confidential mode
    Proof
    Remote attestation

  3. Nothing kept

    The enclave never writes a prompt to disk. Your history stays on your device, under a key only you hold.

    Prompts on disk
    0
    Thinkr sees
    Ciphertext
    History
    Your device

What Thinkr can see

  • Time of a request
  • Size of a request
  • The model you chose
  • Your IP address

What stays sealed

  • Your messages
  • The answers
  • Your files
  • Your history

See how privacy works

Models

The best models, every one labeled.

Open-weight models run in sealed hardware by default. The rest run with zero data retention, and every label says who can read the chat.

  • GPT-OSS
  • DeepSeek
  • Kimi
  • GLM
  • Gemma
  • Llama
  • Claude
  • GPT
  • Gemini
  • Grok
  • GPT-OSS 120B
    Lane
    Private · TEE
    Context
    131K
    Can
    Thinking
    Plan
    Free
  • DeepSeek V4.1 Flash
    Lane
    Private · TEE
    Context
    1M
    Can
    Images · Thinking
    Plan
    Pro
  • Kimi K3
    Lane
    Private · TEE
    Context
    256K
    Can
    Images · Thinking
    Plan
    Pro
  • GLM-5.3
    Lane
    Private · TEE
    Context
    1M
    Can
    Thinking
    Plan
    Credit
  • Gemma 4 31B
    Lane
    Private · TEE
    Context
    256K
    Can
    Images · Thinking
    Plan
    Credit
  • Llama 3.3 70B
    Lane
    Private · TEE
    Context
    128K
    Can
    Text
    Plan
    Credit
  • Claude Opus 5.5
    Lane
    Open · ZDR
    Context
    1M
    Can
    Images · Thinking
    Plan
    Credit
  • Claude Haiku 5.5
    Lane
    Open · ZDR
    Context
    1M
    Can
    Images · Thinking
    Plan
    Pro
  • GPT-6.1 Sol
    Lane
    Open · ZDR
    Context
    1M
    Can
    Images · Thinking
    Plan
    Credit
  • GPT-6 Luna
    Lane
    Open · ZDR
    Context
    1M
    Can
    Images · Thinking
    Plan
    Pro
  • Gemini 3.8 Flash
    Lane
    Open · ZDR
    Context
    1M
    Can
    Images · Thinking
    Plan
    Credit
  • Grok 4.7
    Lane
    Open · ZDR
    Context
    500K
    Can
    Images · Thinking
    Plan
    Credit

Private laneDefault

Sealed hardware

Open-weight models answer inside sealed hardware. Not even Thinkr can read the chat.

Open laneBy choice

Zero data retention

Claude, GPT, Gemini, and Grok run with zero data retention on every request. Their label says that the provider can read the chat.

See plans and prices

By design

Zero is the only number we keep.

No trackers, no ads, no training on your chats, and no chats on our servers.

  • 0Trackers
  • 0Ads
  • 0Chats used for training
  • 0Chats kept on our servers

Thinkr

Think in private.

Your thoughts stay yours: sealed from the first word, labeled for every model, and zero by design.

Start a private chatFollow Thinkr on X