Getting Started
Quickstart
Get Orqen working in your OpenAI-compatible agent in under 5 minutes. You only need to change one line.
1.Create your account
Go to dash.orqen.app/signup and create a free account. Your trial includes 1M saved tokens, with a 75k daily cap and 500k weekly cap — no credit card required.
After signup, you'll see a welcome screen with your first Orqen API key (format: sk-orq-...). Copy it — it's shown once.
2.Connect a provider
Orqen forwards requests to your own LLM provider. In the dashboard, go to Providers and add your OpenAI, Anthropic, or other key. Keys are encrypted before storage and only decrypted per-request.
If you prefer, you can skip this step and pass your provider credentials directly in the request (see Authentication).
3.Change one line in your agent
Point your existing OpenAI SDK client at Orqen's endpoint:
from openai import OpenAI
# Before: direct to OpenAI
# client = OpenAI(api_key="sk-...")
# After: via Orqen (one line changes)
client = OpenAI(
api_key="sk-orq-YOUR_KEY", # Your Orqen key from the dashboard
base_url="https://api.orqen.app/v1",
)
# Your agent code is IDENTICAL from here — nothing changes
response = client.chat.completions.create(
model="gpt-4o",
messages=[{"role": "user", "content": "What is the weather in London?"}],
tools=[
# ... your full tool list — Orqen will prune the irrelevant ones
],
)
print(response.choices[0].message.content)Replace sk-orq-YOUR_KEY with the key from your dashboard. Replace gpt-4o with whichever model you're using. Using native Anthropic or Bedrock tool calls? See provider migration examples.
4.Verify it's working
Run your agent. After the first request, check the dashboard — you should see the request in the Usage tab with the tool count reduced.
Every response includes headers showing what happened:
x-orqen-tools-input: 32 # tools you sent
x-orqen-tools-output: 8 # tools forwarded to the LLM
x-orqen-prune-ratio: 8/32 # output/input
x-orqen-routing: semantic # how Orqen selected the toolsOptional: let Orqen pick the model too
Instead of specifying model="gpt-4o", you can use model="orqen/auto" and Orqen will pick the best model from your connected providers based on task complexity. See Model routing.