Connect to the Prefill Systems API

Base URL

All endpoints below are relative to this base. The same origin serves this website and the API.

1 · Get access

Access is invitation-only during the pilot. Once an administrator invites you, open the invitation link, set a password, and sign in to the dashboard.

2 · Create an API key

In the dashboard → API keys, create a key and give it a name you will recognize in usage records. The raw key is shown once — copy it somewhere safe, because only a hashed form is stored and it can never be retrieved again. You can create several keys (for example one per environment) and revoke any of them at any time. Revocation takes effect immediately.

3 · Make a request

Non-streaming chat completion:

curl https://prefillsystems.com/v1/chat/completions \
  -H "Authorization: Bearer YOUR_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "qwen/qwen3.8-27b",
    "messages": [
      {"role": "user", "content": "Say hello in one short sentence."}
    ]
  }'

Streaming (SSE) — add "stream": true and use -N so tokens arrive as they are generated:

curl -N https://prefillsystems.com/v1/chat/completions \
  -H "Authorization: Bearer YOUR_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "qwen/qwen3.8-27b",
    "stream": true,
    "messages": [
      {"role": "user", "content": "Write a haiku about tokens."}
    ]
  }'

Python with the official OpenAI SDK — point it at our base URL:

from openai import OpenAI

client = OpenAI(
    base_url="https://prefillsystems.com/v1",
    api_key="YOUR_API_KEY",   # replace with your dashboard key
)

response = client.chat.completions.create(
    model="qwen/qwen3.8-27b",
    messages=[{"role": "user", "content": "Explain prefill in one sentence."}],
)

print(response.choices[0].message.content)
# usage is reported per request:
print(response.usage)

Token usage

Every response reports usage with prompt_tokens and completion_tokens; cached prompt tokens are reported separately when the backend provides them. Your dashboard shows the same numbers from gateway-side accounting.

Errors

Errors are JSON with a human-readable detail field:

Operational notes

• Zero data retention: no persistent prompt or completion storage. Transient in-memory prompt caching supports efficient inference. See the data policy.
• Metadata accounting: request id, timestamps, model, token counts, status, timing, and charge are recorded per request for your usage view and for operations.
• Keys are per-account. A key can only be used by the account that created it, and usage is attributed to that account and key.
• No SLA. This is an invitation-only pilot; if you depend on availability, monitor the status page or wait for the production service.
• OpenRouter: once our OpenRouter integration is approved and live, consumers will reach the same gateway through OpenRouter. That channel is planned, not yet available.

Data policy & legal

Read the data policy, privacy policy, and terms of service for details.