Documentation

Using AMP

A practical guide to creating an account, pointing Subquill at AMP, and how billing and privacy work. For the product overview, see the AMP page.

Overview

Subquill AMP is a hosted OpenAI-compatible gateway operated by the Subquill team. One base URL and one API key give you access to many model families, billed per use from a prepaid balance. It speaks the OpenAI chat format, so it slots straight into the endpoint slot Subquill already has.

AMP is entirely optional — Subquill works with any OpenAI-compatible endpoint you already have.

Getting started

Setting AMP up takes four steps:

Create an account in the AMP console and add a prepaid balance
Create an API key in the console
In Subquill's onboarding or Settings, set the base URL to the AMP endpoint below
Pick any model name AMP offers and paste your API key
Run Test connection — it should pass immediately

You can switch between AMP and your own provider any time in Settings; the connection test verifies whichever endpoint is configured.

Configuration

In Subquill, AMP is configured exactly like any other OpenAI-compatible endpoint:

Setting	Value
Base URL	https://api.surfacecyber.com/v1
Model name	Any model listed in the AMP console
API key	An AMP key from the console (starts with sk-)

Pricing

AMP bills from a prepaid balance at per-token rates. The table below lists the indicative rate for each model family; the console shows the full, always-current price list.

Rates are charged per 1M tokens, separately for input and output. There is no subscription and no minimum spend.

Model	Input · $/1M	Output · $/1M	Context
Claude Sonnet	$3	$15	200K
GPT	$1.25	$10	400K
Gemini Flash	$0.30	$2.50	1M
DeepSeek V3	$0.28	$0.42	128K
Qwen3 Max	$1.20	$6	256K
GLM	$0.60	$2.20	200K

Billing

AMP uses a prepaid balance: top up in the console, then each request is billed per token at the provider's rate. The console shows your balance, per-model usage and request history in real time. There is no subscription, and you can stop using AMP at any time — unused balance stays in your account.

Privacy & network

Transcript chunks go from your machine, through the gateway, to the model you selected
Prompts and responses are not retained and never used for model training
Usage is logged only to meter your balance
The Subquill app itself still talks to no backend of its own