The frontier super-intelligence lab

Solving AI privacy, safety, verifiability, and sovereignty

Privacy by default to build confidently

Privacy and verification options available whether you call the models through one API, build in your browser without writing code, or download the weights and run them yourself.

Your request
Private and verifiable
Trust mode per request
Private
Verified
Standard
Response
Trust per request
0G Private API
One endpoint for the whole model catalogue, with a trust mode you set per call: sealed enclave, attested routing, or straight through. Nothing retained, on any plan.
Best for matching protection to the task.
Describe your app
Private and verifiable
0G App interface
Design
Build
Deploy
Created app launched
No code, same guarantees
0G App
Describe an app and launch it from the browser. The inference underneath runs in the same sealed enclaves as the API, so what you build inherits the privacy instead of trading it away.
Best for shipping apps without a codebase.
Your task
Private and verifiable
Work checked as it writes
0GM-1.0
Reviewer
Finished work
Models we built
0GM Models
They do the work rather than describe how to do it. Quality holds from the first step to the last, because a separate model reviews each one as it's written. It costs a fraction of frontier models.
Best for getting it right first time.

Shaping the future of trust modes

Your choice of privacy and proof. One API and no tradeoffs on price or performance.

View all the models

Private, enclave sealed
9 models
The prompt is encrypted end-to-end and opens only in hardware. Zero data retention, while staying cryptographically verifiable. No one can see your data, not even us.
Best for sensitive data
- 0GM-1.0-35B-A3B
- 0GM-1.0-35B-A3B-SIA
- Whisper Large v3
- Qwen3 Embedding 8B
- GLM-5.3
- Laya
- Vela 2.0 9B
- Z-Image-Turbo
Verified, route attested
66 models
Every single request carries proof of which model answered. Published onchain and checkable against the hardware vendor's root of trust.
Best for apps that need provenance
- Qwen3.7 Embedding
- 0GM-1.0-35B-A3B
- 0GM-1.0-35B-A3B-SIA
- Qwen3 Embedding 8B
- GLM-5.3
- Laya
- Vela 2.0 9B
- Whisper Large v3
Standard routing
52 models
The frontier models you already build with, reachable through the same API and the same billing as everything else on 0G. Best price and availability across the full catalog.
Best for everyday tasks
- MiniMax-M3.1-Flash
- DeepSeek-V4.1-Flash
- Qwen3.8-Max
- GLM-5.3-Flash
- Claude Opus 5.5
- GPT-6.1 Sol
- Gemini 3.6 Flash
- Kimi-K3

38M+Requests served
672BTokens processed
140Models
0Bytes of data retained

True privacy from prompt to answer

Your prompt is encrypted before the router sees it. In Private mode it stays sealed all the way through inference. Every step is verifiable.

E2EE Pipeline

Client
Prompt / Response
TEE Encryption
Prompt TEE service
Encrypts · Decrypts · 0 Access
Encrypted response
Router
Blind relay
Forwards · Ciphertext only
TEE Broker
Decrypt · Re-Encrypt · 0 Access

TEE ML
- TEE Attestation · Invisible to all
  Provider never sees data
Partners
- ZDR Partners · TEETLS
  0 retention
- Official APIs · TEETLS
  Alibaba · Zhipu · MiniMax
- Vetted Resellers · Unlisted
  Endpoints not disclosed
SPML
- Post-Hoc Verified
  Idle-GPU network · Lowest cost

A request encrypted by a TEE service, relayed blind by the router and decrypted only in the TEE broker, which routes it to TEE ML, zero-retention partners or SPML

TEE encryption
Sealed before it moves
Your prompt is encrypted inside a secure enclave before it reaches the router, so it can’t be read in transit.
Router + TEE broker
Delivered, never read
The best model is selected for your price and privacy level, and handed off inside a secure enclave. 0G never sees your prompt.
TEE ML
Hidden even from the host
Inference runs in a hardware enclave that the provider’s own machine can’t see into, and the attestation proves it.

Everything for your security team

How we protect your data, and every certification and policy, current and upcoming.

Your Data
- Nothing keptPrompts and responses are not stored, by default and by contract, on every plan.
- Sealed processingInference runs inside hardware enclaves. 0G and cloud operators cannot read it.
- Where it runsInference runs on bare metal in Japan. Account data and logs held in the USA.
UPCOMING
- Region pinningKeep compute and storage in a chosen country.
- Retention windowsSet how long files, chat history and agent memory are kept.
- Verifiable deletionA deletion record you can check, for GDPR and CCPA requests.
AI Safety
- Guardrails before the model runsRules on data, models, spend and outputs block or hold a request before it reaches the model.
- Private models onlyOne switch keeps the whole organization on models that run in sealed compute.
- Human reviewOutputs that leave the company wait for an approver.
UPCOMING
- Agent controlsEach agent gets its own tools, data access and budget, and a kill switch.
- Security testingAutomated attack scans on every agent, with fixes applied as guardrails.
- Risk classificationAgents mapped to EU AI Act tiers automatically.
Proof
- Receipt on every requestA hardware-signed record of which enclave served your request, and that nothing was kept.
- Verify it yourselfRun one command against our gateway and check it against Intel's signature, without trusting 0G.
- Security documentationArchitecture, data flows and the trust model are published.
UPCOMING
- Audit log and evidence packsEvery request, agent step and policy change, exported as a signed pack an auditor checks offline.
- SOC 2 Type II and ISO 27001In progress.
- BAAUnder review for healthcare customers.