Connecting your clients to Quanta

API reference

Every path below lives on https://api.quantarelay.com; all three protocols share one key and one balance.

Endpoint map

Headers and authentication

• Send the key as Authorization: Bearer on OpenAI routes, x-api-key on Anthropic routes and x-goog-api-key on Gemini routes.
• When an Anthropic request carries no anthropic-version, Quanta adds 2023-06-01.
• The key may also travel as a ?key= query parameter, mainly for Gemini clients.
• Your client IP, cookies and CDN headers are never passed upstream.

Limits and timeouts

• Per key: up to 120 requests in any 60 seconds
• Per account: up to 8 requests in flight
• Request body: 25 MB at most
• Per account: up to 20 live keys
• Streams: no total time cap; only 300 s of upstream silence ends one

Error format and status codes

Errors always come back in the OpenAI envelope:

{"error": {"message": "Your Quanta balance is used up. Top up from the dashboard to resume metered calls.", "type": "insufficient_quota", "code": 402}}

Integration guides

With the protocol covered above, here is the setup for each kind of client.

OpenAI SDKs, Codex and other compatible clients

Set base_url to the base address plus /v1 and api_key to your key. Chat goes to /v1/chat/completions; the Responses API is at /v1/responses.

base_url = "https://api.quantarelay.com/v1"   # Quanta, OpenAI-compatible
api_key  = "qnt-your-key-here"

Python, using the official openai package:

from openai import OpenAI  # pip install openai
qt = OpenAI(base_url="https://api.quantarelay.com/v1", api_key="qnt-your-key-here")
qt_reply = qt.chat.completions.create(model="gpt-6.1-sol", messages=[{"role": "user", "content": "Summarise this quarter’s incident log in three bullet points"}])
print(qt_reply.choices[0].message.content)

curl https://api.quantarelay.com/v1/chat/completions \
  -H "Authorization: Bearer qnt-your-key-here" -H "Content-Type: application/json" \
  -d '{"model":"gpt-6.1-sol","messages":[{"role":"user","content":"Summarise this quarter’s incident log in three bullet points"}]}'

GPT ids we suggest starting with:

• gpt-6.1-sol
• gpt-6-astra
• gpt-5.6-sol

Gemini CLI and the native Gemini API

Set GOOGLE_GEMINI_BASE_URL to the bare host (do not append /v1beta) and GEMINI_API_KEY to your key, then run gemini.

export GOOGLE_GEMINI_BASE_URL=https://api.quantarelay.com
export GEMINI_API_KEY=qnt-your-key-here
gemini

curl "https://api.quantarelay.com/v1beta/models/gemini-3.8-flash:generateContent" \
  -H "x-goog-api-key: qnt-your-key-here" -H "Content-Type: application/json" \
  -d '{"contents":[{"role":"user","parts":[{"text":"List three ways to cut token spend on batch jobs"}]}]}'

Gemini models speak only their native protocol: :generateContent for a single answer, :streamGenerateContent?alt=sse for streaming. They cannot be called through /v1/chat/completions.

Gemini model IDs on offer:

• gemini-3.1-pro
• gemini-4-argon
• gemini-3.8-flash
• gemini-2.5-flash

Claude Code and Claude Desktop

Set two environment variables, then run claude: ANTHROPIC_BASE_URL points at the base address (no /v1) and ANTHROPIC_AUTH_TOKEN carries your key.

export ANTHROPIC_BASE_URL=https://api.quantarelay.com
export ANTHROPIC_AUTH_TOKEN=qnt-your-key-here
claude

Claude models are served only to the official Claude Code and Claude Desktop clients, and Claude Code must be version 2.1.285 or newer — an upstream restriction. GPT, Gemini and the rest work from any compatible client or SDK.

Sign up