Docs · Voltspan

1Get access

Voltspan accounts are set up with your team. Contact us and we’ll create your organization, connect single sign-on if you use it, and send you:

Your gateway endpoint, in our cloud or inside your VPC
API keys for each environment, such as staging and production
Access to the console for the people you choose

2Connect your agent

The gateway speaks the OpenAI-compatible API your agent most likely uses already. Change the base URL and key, and keep the model you use today. Nothing else in your agent changes.

Python

import os
from openai import OpenAI

client = OpenAI(
    base_url=os.environ["VOLTSPAN_BASE_URL"],  # your gateway endpoint
    api_key=os.environ["VOLTSPAN_API_KEY"],
)

response = client.chat.completions.create(
    model="your-current-model",  # keep the model you use today
    messages=[{"role": "user", "content": "Summarize this ticket."}],
)
print(response.choices[0].message.content)

curl

curl "$VOLTSPAN_BASE_URL/chat/completions" \
  -H "Authorization: Bearer $VOLTSPAN_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"model": "your-current-model", "messages": [{"role": "user", "content": "Hello"}]}'

Tool calls, streaming and multi-turn sessions pass through unchanged.

3See every run

From the first request, every model call, tool call and result is captured as a trace in storage you control. In the console you can search runs, replay them step by step, and see where and why they fail.

About VS Data →

4Compare models on your own tasks

Pick a set of real runs and turn it into an eval. Run it against the model you use today, open models, and later your own, and compare quality, latency and cost side by side before you change anything in production.

5Train your own model

When you have enough good runs, we help you train an open model on them: first supervised fine-tuning on your traces, then reinforcement learning in sandboxes, where the model tries your tasks and is rewarded for getting them right. Your data trains only your model, and the weights are yours.

How training works →

6Serve it

Your model is served behind the same gateway. Shift a share of traffic to it, watch the evals, and roll back in one step if anything looks off.

Python

response = client.chat.completions.create(
    model="your-org/your-model",  # the model you trained
    messages=messages,
)

Where it runs

Everything above runs the same way in Voltspan Cloud, in your VPC, or in your own data center. Only the endpoint changes.

Security and deployment →

Get help

Your account comes with engineers who know your setup. Reach them through your shared channel, or contact us. The full API reference is shared with your team during onboarding.