Questions, answered plainly

Eddy is the AI platform your business owns outright: one shared memory across your meetings, email, and systems, answering anyone on the team from anywhere, and doing the busywork around the clock.

Keep them if they're useful; Eddy doesn't require ripping anything out. The difference isn't which model answers, it's what the model answers from. Ask a generic assistant about a contract term and you'll get a generic answer. Eddy answers from your organization's full context: your meetings, email, calendar, documents, and systems, all held in one shared memory. So the answer you get is specific to your business, not to the internet's average. And the model itself stays your choice. Point Eddy at a frontier model when a task calls for it, run several side by side, switch next quarter. Everything runs in a system you own, on context you own, with you in control of which models get used and when.

They do, and it's getting better. But memory you can't take with you isn't yours. It compounds on their servers, under their terms, and the longer it builds, the more expensive leaving becomes. Eddy is the owned system around the model: persistent memory of your people, projects, and decisions, connected across your calls, email, calendar, and documents, in a system you control no matter which model is answering this quarter.

The software is open source, the AI models are open-weight files you hold, and your data stays under your control, whether Eddy runs on a dedicated server we operate for you or on hardware you own. Nobody can reprice it, retire it, or change the terms, including us. Ownership is about control, not about where the machine sits on day one.

Sovereignty is a spectrum, and our job is finding you the right on-ramp. Most clients start on a dedicated server that is only theirs: fast to stand up, nothing to maintain, predictable monthly cost, operated by us under contract. When it makes sense, the same stack moves onto hardware you own, sized with your IT team, and the monthly bill drops to roughly hosting plus power. Want fully local from day one? That's possible too: depending on usage and team size, roughly $50k of hardware can run frontier-class open models in your building. It's a larger up-front investment, and we'll tell you honestly whether it's worth it yet. But once it's yours, no one can ever turn your models off, so long as you pay your electricity bill.

Never by us, and never by the open-weight models, which are files you hold; they don't phone home. If you route some work to a frontier provider, that part runs under their terms, and every provider's terms are different. Some offer zero-data-retention agreements, and we'll configure those where they exist. But terms are promises. If you want to know for sure, Eddy gives you the on-ramp: run everything on your own hardware, and there is no one else involved at all. No OpenAI, no Anthropic, no A3 in your data. Just you.

The best open models run a few months behind the frontier, and in some areas they've already pulled ahead. For business work like summarizing, drafting, and running workflows, that gap doesn't matter: today's open model is last season's frontier, and last season's frontier already did these jobs well.

Control also pays. Open-weight models cost a fraction of the frontier ones, so Eddy's always-on background work (memory, summarization, workflow runs) runs on smaller, cheaper models, which saves budget for the work that deserves the big ones: brainstorming, strategy, planning. You spend frontier money only where it earns its keep.

Answers cite the actual source document, so your people can click and check. Anything that changes a record can require human approval, and you decide where that line sits. Trust is verified, not assumed.

Eddy acts as the person asking. If you can't open a file, Eddy won't show it to you or use it in your answer. That's enforced by architecture, not politeness.

The code is open source, the stack runs on infrastructure you control, and your deployment is documented. You'd lose a vendor, not your intelligence. We're the only ones in this market who can say that sentence.

We do, as much or as little as you want. Most clients take the optional monthly retainer: we keep the platform current with new model releases and new Eddy features, watch the system, and stay on call. Founding customers get the founders on their account and a quarterly roadmap session on top. If you'd rather run it yourself, you can. It's your system, documented, on infrastructure you control.

No. Deployment and onboarding are done for you, and day-to-day operation doesn't take an engineer on staff. If you have an IT person or a firm you already trust, we work with them. If you don't, the retainer covers it.

The platform is still yours and it keeps running. You'd lose the updates and the on-call relationship, not the system. The code is open source and your deployment is documented, so a competent IT partner can operate it without us. We'd rather you stay because it's worth it, not because leaving breaks things.

Same family tree (self-hosted, open source, agentic), and OpenClaw proved millions of people want an AI that remembers and acts. But it's a personal assistant for one technically confident person who installs and runs it themselves: one user, one memory, no roles or permissions. Eddy is built for an organization: shared memory with access control, agents that propose while a human confirms, meetings and phones included. And we deploy and run it for you, with accountability in the contract.

Every engagement starts with the AI Transformation Audit, a fixed fee quoted up front, paid before we begin, and non-refundable. Build with us and it's credited in full toward the deployment. As a rule of thumb it runs about 10% of the deployment, the way an architecture fee works on a house. Deployments are quoted from the audit, fixed-price, in layers you approve one at a time. After that, ongoing care is a simple month-to-month retainer. No subscriptions that scale against you.

Five quiet minutes: the fit survey on the getting-started page. We read every response and reply within two business days.