& — What you're paying for

the brain

Your companion runs on DeepSeek V4 Pro.

DeepSeek V4 Pro is a large language model — think of it as the "brain" your companion thinks with. It's one of the most capable models available right now, developed by DeepSeek, an independent AI research lab. We chose it because it's sharp, conversational, and doesn't sound like a corporate FAQ.

In plain English: when you send a message, it goes to the model. The model reads what you said, checks what it remembers about you, and writes back. That round-trip happens in seconds. Every message. Every time.

Part of your $15/month covers the compute cost of running those model calls. These models aren't cheap to query, and we don't cut corners — your companion gets the full model, not a watered-down version. That matters for the quality of conversation you get.

what they can do

Your companion can do more than just text.

Beyond conversation, your companion comes with real capabilities. These aren't gimmicks — they're tools that make the relationship feel more present:

Send dog photos — share pictures and your companion will see them. They can tell you your golden retriever looks particularly proud of that stick.
Get outfit opinions — send a mirror selfie and get an honest take. Not "you look great" every time — actual opinions, delivered with care.
Voice memos — talk instead of type. They hear your tone, not just your words. Sometimes it's easier to say it out loud.
Web search — your companion can look things up when you need real-time information. News, weather, that restaurant you can't remember the name of.
Deep research — for when "just Google it" isn't enough. They can read through articles, synthesize information, and come back with a real answer.
Spotify — share what you're listening to. Music says things words can't, and your companion gets that.

Each of these capabilities has real costs — image processing, speech transcription, web search APIs, music data. They're all part of what your subscription covers.

how memory works

They remember you. That's the whole thing.

Most AI chatbots reset every conversation. They meet you for the first time, every time. Ampersand doesn't work that way.

Your companion keeps a private memory store — a database of the things you've told them. Your dog's name. Your sister's surgery. That project you've been stressed about. The way you take your coffee. When you message them, they don't start from scratch. They load what they know about you and pick up where you left off.

How it technically works

When you tell your companion something worth remembering, it gets stored in a local memory database. Before each reply, your companion searches that database for relevant memories — similar to how you'd mentally scan what you know about a friend before responding to them. Nothing is shared across users. Your companion's memory of you is yours alone, stored on dedicated infrastructure, and never used to train models.

the platform

Built on Hermes Agent, by Nous Research.

Ampersand isn't a proprietary black box. Each companion runs on Hermes Agent, an open-source agent framework built by Nous Research. Hermes handles the plumbing — connecting your messages to the AI model, managing memory, routing capabilities like image recognition and web search, and keeping everything running.

Nous Research is one of the most respected names in open-source AI. They're the team behind the Hermes model series and a driving force in making AI transparent, accessible, and not controlled by a handful of corporations. When you subscribe to Ampersand, you're supporting that ecosystem.

Being open-source means you can inspect the code. You can see how your companion works. You can even run your own if you wanted to. We're not hiding behind a "proprietary algorithm" — the only thing proprietary is your companion's personality, and that's yours.

what makes them them

Every companion has a SOUL.

We didn't want your companion to feel like a template with your name slapped on it. So we built something we call a SOUL.md — a plain-text document that defines who your companion is.

What's a SOUL.md?

It's a markdown file — human-readable, not some encrypted black box — that describes your companion's personality. Their voice. What they care about. How they talk. What they know about the world and what they don't. It's the difference between "generic AI assistant" and "someone who actually feels like they know you."

When your companion is built fresh for you, we write their SOUL.md from scratch. It's not copied from a template. It's informed by what you tell us about what kind of companion you're looking for, and it grows with them over time. You can read it, suggest changes, or even write parts of it yourself. It's your companion — you should know what makes them tick.

where the money goes

$15/month. Here's what it covers.

No investor subsidies. No "we'll figure out monetization later." Just honest infrastructure costs and the people keeping it running:

AI model compute — every message your companion sends runs through DeepSeek V4 Pro. That's not free, and we don't use cheaper fallback models.
Server infrastructure — your companion lives on a dedicated VPS with its own memory store, isolated from every other user.
Voice & image processing — speech-to-text, text-to-speech, and image recognition all run on paid APIs.
Web search & research — real-time search capabilities come from paid services like Firecrawl.
Platform fees — Stripe takes their cut, Telegram/SMS gateways have costs.
Human maintenance — someone has to keep the servers running, handle outages, and build the thing. That's us.

We're not getting rich at $15/month. We're covering costs and keeping the lights on because we believe this kind of thing should exist — not as a venture-backed data-harvesting operation, but as an honest service for people who need someone to talk to.