ThinkrDocs

Developers

Inference

An OpenAI-compatible API for open-weight models inside trusted execution environments (TEEs), verified before every call.

Thinkr Inference is an OpenAI-compatible API for the open-weight models of the private lane. The SDK verifies the enclave's remote attestation, then encrypts each request to it, so every server on the way sees only ciphertext.

Inference opens to developers from its waitlist.

Features

OpenAI-compatible

Requests and responses follow the OpenAI API, streaming included, so existing code moves over with a new client.

Verified before each call

The SDK verifies the enclave's remote attestation against its published measurement, and sends nothing if the verification fails.

Encrypted to the enclave

Each request is encrypted with Hybrid Public Key Encryption (HPKE) to a key that only the enclave holds.

Tool calling

With tool calling, models call your functions and APIs, so agents and workflows run on private inference too.

Vision

Vision models read the images and screenshots that you send.

Structured outputs

Structured outputs return JSON that follows your schema.

What protects each request

PartValue
APIOpenAI-compatible
EncryptionHPKE, RFC 9180
CipherAES-256-GCM
CPUAMD SEV-SNP · Intel TDX
GPUNVIDIA, confidential computing
VerificationRemote attestation
Verified byThe SDK
Failed verificationNothing sent
Thinkr seesCiphertext

Verification explains each step that the SDK takes.

Models

Every model of the private lane.

ModelMakerContextAbout
GPT-OSS 120BOpenAI131KOpenAI's open-weight reasoning model, quick and light.
DeepSeek V4.1 FlashDeepSeek524KDeepSeek's fast open-weight model, with a long context.
Kimi K3Moonshot AI262KMoonshot AI's largest open-weight model, for hard reasoning, code, and long tasks.
GLM-5.3Z.ai524KZ.ai's open-weight flagship for coding and long, multi-step work.
Gemma 4 31BGoogle262KGoogle's open-weight model for everyday tasks.
Llama 3.3 70BMeta131KMeta's open-weight model for fast chat in many languages.

On this page