Your first API request.

Open API key management · Beta credit policy

1. Set your API key

Copy

export SUBFOR_API_KEY="your-own-api-key"
export SUBFOR_BASE_URL="https://token.subfor.me/v1"

Enter the real key only in your terminal. Keep it out of source control and screenshots.

2. List available models

Copy

curl "$SUBFOR_BASE_URL/models" \
  -H "Authorization: Bearer $SUBFOR_API_KEY"

3. Make a chat request

Copy

curl "$SUBFOR_BASE_URL/chat/completions" \
  -H "Authorization: Bearer $SUBFOR_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"model":"gemini-3.5-flash-lite","messages":[{"role":"user","content":"Hello!"}],"max_tokens":128}'

4. Stream output and usage

Copy

curl -N "$SUBFOR_BASE_URL/chat/completions" \
  -H "Authorization: Bearer $SUBFOR_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"model":"gemini-3.5-flash-lite","messages":[{"role":"user","content":"Count from 1 to 10."}],"max_tokens":128,"stream":true,"stream_options":{"include_usage":true}}'

Text arrives in SSE data events; usage normally arrives near the end. Heartbeats are not output tokens. Cancellation can still incur usage for work already performed.

5. Python SDK

pip install openai

Copy

import os
from openai import OpenAI

client = OpenAI(
    api_key=os.environ["SUBFOR_API_KEY"],
    base_url="https://token.subfor.me/v1",
)
response = client.chat.completions.create(
    model="gemini-3.5-flash-lite",
    messages=[{"role": "user", "content": "Hello!"}],
    max_tokens=128,
)
print(response.choices[0].message.content)

Tool calls and client compatibility

Gemini Flash Lite Chat Completions tool selection and tool-result continuation have been verified. Preserve tool_call_id and return the result with role: "tool".

Select an OpenAI-compatible provider in your client and enter the base URL above. Full client, Responses and WebSocket compatibility is not promised. A dedicated Codex++ integration will follow separate validation.

Troubleshooting

• 401: Check for an invalid, expired or revoked key.
• 403 / insufficient quota: Check account balance, key budget and model access. New accounts need an allocation from their inviter.
• 429: Reduce concurrency, back off and respect Retry-After.
• 5xx / timeout: Check the performance page. Do not automatically retry tools or payment actions that may already have executed.