Things people actually ask.

Your agent runs on a machine you control (laptop, office server, private VPS), not on ours. Your data flows from your machine to the AI provider you've chosen, using your API key. We never see it. We never store it. We never train on it.

During the build, we see only the samples you choose to share: example emails, sample contracts, things you've redacted. Once the agent is live on your machine, your live data never touches anything we own. We've architected it so we couldn't see it even if we wanted to.

Yes, your data goes from your machine to the model provider (OpenAI, Anthropic, Google, etc.) so they can generate a response. That's unavoidable for cloud LLMs. But: (1) it goes directly, not through us; (2) reputable providers don't train on API inputs by default; (3) if you want zero data leaving your hardware, we deploy with an open model (Llama, Qwen, Mistral) running locally instead.

Yes. For regulated industries or paranoid setups, we deploy with an open-source model running on your hardware. Quality is slightly behind the frontier models, but it's improving fast, and the data never leaves your building.

Bring Your Own Key. You sign up with OpenAI, Anthropic, Google (any provider), and your agent connects to them directly. No hidden platform meter from us. If we pick the wrong provider, switch. If a better model lands tomorrow, switch. You're never locked in.

A solo founder needs different things from a 40-person team. We size the engagement to your time-sinks, your team, and your tools, after a free 30-minute discovery call. You get a written scope and a clear timeline before anything starts. No surprise additions. No upselling.

Two things, both modest. (1) AI provider tokens, billed to you directly by your chosen provider, usage-based, so it varies with how much your agents work. (2) Optional support if you want us watching the system with you. No plan ladder. No seat maze. The agent runs on hardware you already own.

Never. You pay OpenAI, Anthropic, Google (whomever you choose) directly. We never see the bill. This is non-negotiable for us: the moment a vendor takes a margin on tokens, they're incentivised to keep you using more of them.

No. We build. You describe the work that wastes your time, and we hand you back an agent that does it. Talk to it like an assistant. If you want to tinker, we hand you the keys too.

Yes, and this is where it gets useful. Contracts. SOPs. Past emails. Product specs. Your style guide. Your agent learns from it all, and answers like someone who's been at your company for years. The data stays on your machine.

It will, sometimes. Agents flag what they're unsure about and ask for human review. You set what they can do alone and what needs your nod first. The optional Support layer means we tune drift out before you notice it.

Most agents are made and deployed in under 4 hours, often the same afternoon as our first call. A full transformation engagement runs twelve weeks, end to end.

A modern laptop (M-series Mac or recent Windows / Linux machine) handles most agents. For team deployments, a basic VPS or your office server is enough. For local open-source models, you need a machine with 16GB+ RAM and ideally a GPU. We'll tell you exactly during the audit.

You keep everything. The agent is already on your machine. The prompts, the configuration, the documentation: all yours. Your data was always yours. Walk away whenever. No clawback. Your signed support terms define the handoff.

Your agent keeps running. Honestly. The whole point of "deployed on your machine, BYOK" is that nothing depends on HeyOne AI being alive. The code is yours, the prompts are yours, the keys are yours. The architecture is your insurance against exactly this.

Absolutely, we encourage it. The handbook we give you walks through editing prompts, adding integrations, swapping models. Most customers are independent within 60 days. The Support layer is for when you'd rather not.

Yes. We build agents in a model-agnostic way: swap from OpenAI to Claude, from Gemini to DeepSeek, or to a local model, and the agent keeps working. If you're on Support, we test new models against your workflows and tell you when it's worth switching.