Documentation
Everything you need to get started with OpenTau.
Getting Started
Three steps. No SDK to install, no blockchain wallet to set up, no hoops to jump through.
Get your key
Sign up on the dashboard and get an API key instantly. No wallet, no KYC, no waiting.
Set your endpoint
Point your existing code to api.opentau.ai/v1 instead of the OpenAI endpoint. That is the only change.
Make a request
Use any OpenAI-compatible SDK or library. Python, TypeScript, curl, anything. Same format, same response shape.
# That's it. Use any OpenAI SDK. base_url = "https://api.opentau.ai/v1" api_key = "otau_your-key"
Available Endpoints
OpenTau is fully OpenAI-compatible. If you have code that talks to OpenAI, it already works here.
Chat with any model. Send messages, get responses. Supports streaming.
{ "model": "meta-llama/llama-3.3-70b-instruct",
"messages": [{ "role": "user", "content": "Hello" }] }
Browse all 532+ available models. Returns standard OpenAI model list format.
Search the web, X, and Reddit via decentralized search miners. No tracking, no filter bubbles.
{ "query": "latest Bittensor subnet launches",
"sources": ["web", "twitter"] }
Check if an image is AI-generated or real. Returns a confidence score and analysis.
{ "image_url": "https://example.com/photo.jpg" }
Train a custom model on your data. Upload a JSONL file and get a fine-tuned model back.
{ "model": "meta-llama/llama-3.3-70b-instruct",
"training_file": "file-abc123" }
Store and retrieve files on decentralized storage. Persistent, censorship-resistant, no single point of failure.
{ "action": "upload",
"file": "base64-encoded-data" }
Models
OpenTau gives you access to 532+ models from every major open-source family. All served by decentralized miners competing to give you the fastest, cheapest inference.
Switching models is trivial. Change the model string. Everything else stays exactly the same.
# Just change the model string model = "meta-llama/llama-3.3-70b-instruct" model = "deepseek-ai/deepseek-r1" model = "mistralai/mixtral-8x22b-instruct-v0.1"
Use the full lowercase model name (e.g., meta-llama/llama-3.3-70b-instruct). Hit /v1/models to browse the full list.
Authentication
Every request needs an Authorization header with your API key. That is the only authentication.
Authorization: Bearer otau_your-api-key
Your key is shown in the dashboard right after signup. Keep it secret. If it leaks, rotate it from the dashboard.
Rate limits
Free tier: 1,000 requests per day. No deposit needed.
Paid: based on your credit balance. No hard cap.
Billing & Credits
OpenTau uses a simple pay-as-you-go credit system. No subscriptions, no monthly commitments, no surprises.
Free tier
1,000 requests per day. No deposit, no credit card, no wallet. Just sign up and start calling the API.
Pay as you go
Deposit USDC on Base. Every request deducts a tiny amount from your balance. Roughly ~$0.0001 per request, varying by model size. Credits appear within ~30 seconds of deposit.
When credits run out
The API returns a 402 Payment Required response. Top up anytime from the dashboard and you are back instantly.
See the full breakdown on the Pricing page.
Cross-Subnet Pipelines
OpenTau is the only API that lets you chain multiple decentralized AI services in a single call. This is possible because we route across Bittensor's specialized subnets, not just one inference provider.
A few things you can do that no centralized API offers:
Full examples and pipeline syntax on the Pipelines page.
SDKs & Integration
You do not need a new library. OpenTau works with any OpenAI-compatible SDK because the API format is identical. Just change the base URL.
from openai import OpenAI client = OpenAI( base_url="https://api.opentau.ai/v1", api_key="otau_your-key" )
curl https://api.opentau.ai/v1/chat/completions \ -H "Authorization: Bearer otau_your-key" \ -d '{"model":"meta-llama/llama-3.3-70b-instruct", "messages":[{"role":"user","content":"Hi"}]}'
Compatible with frameworks you already use:
Frequently Asked
Yes. The free tier gives you 1,000 requests per day with no deposit, no credit card, and no wallet. If you need more, deposit USDC on Base and pay roughly $0.0001 per request. There are no hidden fees.
OpenRouter routes to centralized providers (OpenAI, Anthropic, Google). OpenTau routes to decentralized miners on Bittensor. No single company controls the inference. Plus, we offer services OpenRouter cannot: decentralized search, fine-tuning, storage, image verification, and cross-subnet pipelines.
Your prompts are forwarded to miners who process them and return results. OpenTau does not log or store your prompts. Miners on Bittensor are incentivized to respond correctly, not to harvest data. For maximum privacy, use the confidential compute subnet which processes data in secure enclaves.
Nothing, from your perspective. The gateway automatically routes to the next available miner. There are hundreds of miners competing on each subnet, so there is always redundancy. You will never get a "service unavailable" because one machine went offline.
Absolutely. OpenTau is designed for autonomous agents. It works with ElizaOS, Virtuals GAME SDK, CrewAI, AutoGPT, and any framework that speaks the OpenAI format. Many agents on Bittensor already use us as their backbone.
Many open-source models available through OpenTau have minimal or no content filtering. Decentralized miners run the models as-is, without adding extra safety layers. Check the model list for specific uncensored variants. This is one of the key advantages of decentralized inference.
Not for the free tier. Just sign up with an email and start making requests. You only need a wallet if you want to deposit USDC on Base for paid usage. Even then, the flow is simple: send USDC to the deposit address shown in your dashboard.
Reach out on X or join the community. We are responsive and actively building. If something is broken or confusing, we want to know about it.