HammerAI

We opened a character from the front page with no cookie, no account and no email. One modal stands in the way: an adult confirmation that also asks you to accept the terms, cleared with a single button. After that the chat is live.

We sent one plain question and got a coherent two-sentence answer. The request went to an endpoint called /api/cloud/chat-anon, the header named the model, and a notice confirmed the chat had been saved locally, in the browser rather than on an account.

That is more than almost any platform we cover lets a stranger do. It is also one exchange on the free cloud model, and the reason this page carries no score. The toolbar under the thread shows what the product is for: Impersonate, Author's Note, Dice, Continue. These are writing tools for roleplay, not companion features.

Image generation is the wall. The generator loads, accepts a prompt, and keeps its button greyed out with a tooltip asking you to log in. The pricing page explains why: images are paid for in credits, and the free plan has none.

HammerAI publishes every price on one open page, which already puts it ahead of much of this category. Paid plans buy three things: unlimited cloud messages, a longer memory, and credits for images and text-to-speech.

Reading the pricing page twice, we found it contradicts itself, and the rest of the site adds to the list. None of these is dramatic on its own. Together they mean that no single HammerAI page can be taken at its word, including the one you pay from.

The desktop app is the reason HammerAI gets recommended. Its page says it packages Ollama and llama.cpp, needs no login, runs offline, and keeps messages on your computer, deleting them when you close it. You can also point it at your own Ollama model.

What we could check from outside: the download buttons lead to a public GitHub repository that holds releases only, with no source code. The Linux Flatpak carries an Electron package name, so the app is built on Electron, a browser engine packaged as a desktop program. Being closed source is common. It does mean the claim that nothing leaves your machine rests on HammerAI's word, which you can test yourself by disconnecting from the internet, as their own FAQ suggests.

We did not install it, so we cannot tell you how fast a local model answers or how good it is. That depends on your graphics card far more than on HammerAI. Anything outside local models, cloud chat, images and voice, goes through their servers even from the desktop app.

On the pages we loaded, the network log is short. There is no advertising pixel, no session recorder, no social tracker.

Nowhere. The footer, the terms and the privacy policy all say HammerAI and nothing more. There is no company name and no address. The pricing page adds that the product is built by one developer, and the only contact is team@hammerai.com.

The terms still choose a venue: Delaware law, and binding arbitration under the American Arbitration Association, which you can opt out of by email within thirty days of accepting them. An arbitration clause without a named party to arbitrate against is an odd combination. It matters little for a free user and more for someone buying the larger credit packs with no refund.

The robots.txt is more thoughtful than the legal pages. It lets search engines in, keeps private chat URLs out, shuts the main SEO crawlers out of character pages, and declares that the content is not to be used for AI training.

We score platforms on six criteria. Two of them we could not measure without an account or a payment. Image generation is closed to anonymous visitors. Chat quality we saw on one free cloud model only, while the paid models, the longer memory and the local models in the desktop app, which are the actual case for HammerAI, stayed untested.

We would rather leave it unscored than guess at two of six. That is why HammerAI is absent from our roleplay chatbot ranking and from the comparison tables. When those two criteria can be checked, it gets a score like everything else.