Quickstart
Get OrchVynt running locally in under 10 minutes. You'll pull the Docker image, write a minimal routing config, and route your first agent invocation through the control plane.
Prerequisites
- Docker installed and running
- An API key for at least one LLM provider (OpenAI, Anthropic, or Mistral)
- An OrchVynt API key (sign up for early access to get one)
Step 1: Pull the Docker image
terminal
$ docker pull orchvynt/control-plane:latest
Status: Downloaded newer image for orchvynt/control-plane:latest
Step 2: Write a minimal config
Create an orchvynt.yaml file in your project directory:
orchvynt.yaml
orchvynt:
api_key: ${ORCHVYNT_API_KEY}
listen_port: 4821
providers:
- id: openai
api_key: ${OPENAI_API_KEY}
routing:
default_model: gpt-4o-mini
rules:
- workload: synthesis
model: gpt-4o
telemetry:
enabled: true
Step 3: Start the control plane
terminal
$ docker run -p 4821:4821 \
-v ./orchvynt.yaml:/etc/orchvynt/config.yaml \
-e ORCHVYNT_API_KEY=$ORCHVYNT_API_KEY \
-e OPENAI_API_KEY=$OPENAI_API_KEY \
orchvynt/control-plane:latest
OrchVynt control plane listening on :4821
Step 4: Route your first invocation
Call the OrchVynt routing endpoint instead of the provider API directly:
Python SDK
import orchvynt
client = orchvynt.Client(
base_url="http://localhost:4821",
api_key="ov_key_..."
)
response = client.route(
workload="synthesis",
messages=[{"role": "user", "content": "Summarize this document."}]
)
print(response.content)
# OrchVynt routed to gpt-4o based on workload=synthesis
You're now routing through the OrchVynt control plane. All invocations are telemetered. Add fallback chains, budget rules, or HITL gates by extending
orchvynt.yaml and restarting.