The Chutes Starter Guide

From zero to a working API call on decentralized compute. Every price in this guide is a live-site snapshot as of 2026-08-17; the current numbers are always at chutes.ai/pricing.

Why This Exists

Most AI bills route through a chain of middlemen, and every layer takes a cut before the bill reaches you. Chutes shortens the chain. It runs decentralized, serverless inference for open-source models on Bittensor SN64, where independent GPU operators compete to serve your workload. Competition on the supply side, instead of a markup stack. And 100 percent of served models run inside hardware-attested TEEs with end-to-end encryption, so cheaper does not mean exposed.

This guide gets you from no account to a working API call, with money on the balance and a model picked, in one sitting.

Step 1: Create Your Account

Go to chutes.ai/signup. You can sign up with Google or GitHub, or just pick a username and Chutes mints a fingerprint credential you sign in with. Connecting a Bittensor wallet is optional; you do not need one to use the platform.

Step 2: Create an API Key

Once you are in, create an API key in the chutes.ai app. Keys carry the prefix cpk_ and go in the Authorization header of every request. If you live in the terminal, the chutes CLI can create keys too, scoped as tightly as you want.

Treat the key like a password: environment variable, not source code.

Step 3: Make the Call

The API is OpenAI-compatible. Base URL: llm.chutes.ai/v1. If your code already speaks to an OpenAI-style endpoint, you change the base URL and the model name, and you are done.

import os
import requests

payload = {
    "model": "deepseek-ai/DeepSeek-V4-Flash-0731-TEE",
    "messages": [{"role": "user", "content": "Ship a concise launch checklist."}],
    "stream": True,
}

response = requests.post(
    "https://llm.chutes.ai/v1/chat/completions",
    headers={
        "Authorization": f"Bearer {os.environ['CHUTES_API_KEY']}",
        "Content-Type": "application/json",
    },
    json=payload,
    stream=True,
)

for line in response.iter_lines():
    if line and line.startswith(b"data: "):
        print(line.decode()[6:])

curl https://llm.chutes.ai/v1/chat/completions \
  -H "Authorization: Bearer $CHUTES_API_KEY" \
  -d '{"model": "deepseek-ai/DeepSeek-V4-Flash-0731-TEE",
       "messages": [{"role": "user", "content": "hello"}]}'

The Vercel AI SDK and n8n integrations ship out of the box, and the chutes-agent-toolkit drops all-TEE inference into Claude or any OpenAI-compatible client.

Step 4: Add Money

Top up your balance in the app. Pay-as-you-go has no minimum commitment and no subscription requirement; you pay for the tokens you use. Chutes accepts 25+ payment methods via Stripe, spanning cards, crypto, bank transfers and more, and also takes TAO and Bittensor alpha tokens.

If you prefer a predictable monthly bill, two optional plans exist: Plus at $10 per month (a bundled daily request quota, then 6 percent off pay-as-you-go rates) and Pro at $20 per month (a larger daily quota, then 10 percent off). Enterprise customers can arrange custom terms, including invoicing.

The Catalog: Models and Prices

Prices are per 1M tokens, as of 2026-08-17.

All models run in TEEs. Live list: chutes.ai/pricing and llm.chutes.ai/v1/models.

Which Model Should You Start With?

Four picks, by job:

• Cheapest possible test — Mistral-Nemo-Instruct-2407-TEE or Nemotron-3-Nano-Omni-30B-TEE. Tied at the bottom of the price list; run your routine tasks (summaries, drafts, formatting, extraction) here and watch what it does to the bill.
• Everyday all-rounder — DeepSeek-V4-Flash-0731-TEE. A 1M-token context near the bottom of the price table; a strong default for a real workload.
• Coding and agents — GLM-5.1-TEE. The catalog describes it as optimized for agentic tasks and coding, built for sustained problem-solving with tool use.
• Open-weights flagship — Kimi-K3-TEE. A 1M context with image and video input on top of text, for the workloads you would otherwise send to a closed frontier model.

Prices and the catalog move often. Check chutes.ai/pricing before you commit a workload.

The Privacy Part

Running a cost test should not mean leaking your data. Every model above runs inside a hardware-attested TEE with end-to-end encryption, and the security architecture is auditable: Chutes open-sourced its TEE stack (sek8s) and the encryption layer. You can verify the claims instead of trusting them. On speed, DFLASH delivers 30 to 50 percent real-world decode speedups on tested models.

Start Now

Your first call can be running before your coffee cools. Sign up at chutes.ai and create a key. Point your client at llm.chutes.ai/v1.