One API for the best
open models
AI Root is an OpenAI-compatible gateway to the strongest cheap open models — DeepSeek, Qwen, GLM, Kimi and more — behind one endpoint. Point your apps and agents at a single URL, with automatic failover, response caching, and per-key budget caps.
from openai import OpenAI
client = OpenAI(
base_url="https://api.ai-root.com/v1", # ← the only change
api_key="sk-air_...", # your AI Root key
)
client.chat.completions.create(
model="cheapest", # or auto / fastest / a model name
messages=[{"role": "user", "content": "Hello"}],
)DeepSeek, Qwen, GLM, Kimi and more — the frontier of cheap open models, from $0.10 per million tokens (60–70% below GPT/Claude). One key across every provider, each model labelled by the region it runs in. Route to auto, cheapest, fastest, or a named model.
Automatic failover routes around a provider outage or quota before your app or agent ever sees it — critical for long-running agent loops. Identical repeated calls are served from cache: free, and instant.
Mint a key per app, customer or agent; tag it; cap it. A monthly budget and hard cap mean a runaway agent loop stops itself instead of draining your account. Watch spend per key, export a CSV — one bill, no surprises.
Not sure which model?
Run one prompt across several models at once and compare quality, cost and latency side by side. Pick the cheapest that's good enough — then change one string to ship it.
Know where every request runs
AI Root runs on reliable, hosted infrastructure, and every model is labelled with the region it runs in — so you always know where a request is served. Your account and usage data are stored onshore in Australia. Transparent, predictable, and easy to put in front of real workloads.
Simple, prepaid, no lock-in
Token-metered at published rates — from $0.10 per million tokens. Top up by card, no subscription, no minimum. Per-key caps let you hand a key to any app or agent without budget risk. New accounts start with free trial credit.
Running real traffic across a portfolio of apps, agents or customers? Wholesale rates, priority support, and help mapping keys to your cost centres. Talk to us once the gateway has earned it.
Frequently asked questions
What is AI Root?+
AI Root is an OpenAI-compatible AI gateway that gives your apps and AI agents one endpoint to the best cheap open models — DeepSeek, Qwen, GLM, Kimi, Llama and more. You change one base URL in your existing OpenAI code and keep everything else the same.
Which models does AI Root support?+
The leading open-weight models — DeepSeek, Qwen, GLM, Kimi, Llama and Mixtral — behind one endpoint, each labelled with the region it runs in. Route to 'auto', 'cheapest' or 'fastest', or request a specific model by name.
How much does AI Root cost?+
AI Root is pay-as-you-go from $0.10 per million tokens — roughly 60–70% below GPT-4-class and Claude pricing. There is no subscription and no minimum, and new accounts start with free trial credit.
Is AI Root really OpenAI-compatible?+
Yes. AI Root implements the OpenAI Chat Completions API, so the official OpenAI SDKs and any OpenAI-compatible tool work by changing only the base URL and API key — no rewrite.
Can I use AI Root with AI agents and agent frameworks?+
Yes — agents are a first-class use case. Because AI Root is OpenAI-compatible, agent frameworks like LangChain, CrewAI, LlamaIndex and the OpenAI Agents SDK work by pointing at the AI Root base URL. Give each agent its own key with a monthly budget and a hard cap, so a runaway loop stops itself instead of draining your account, and automatic failover keeps agents running through a provider outage.
Why use AI Root instead of calling a model provider directly?+
One endpoint spans many providers with automatic failover, response caching and per-key cost tracking. Instead of integrating and maintaining several provider accounts and SDKs, you point at one OpenAI-compatible URL and get one bill with a per-app or per-agent cost breakdown.
Where is my data processed?+
AI Root is built and hosted in Australia, and your account and usage data are stored onshore. Every model is labelled with the region it runs in, so you always know where a request is served.
Point your app or agent at it in a minute
One base URL, one key, free trial credit to test with.
Get started →