Bring your own model

Model access is fragmenting, and that is a good thing. Some models you reach through a subscription. Some you reach with an API key and a credit card. And a growing number run on hardware you already own: a gaming PC, a home server, a work laptop with a decent GPU. What has not caught up is the software. Most chat apps assume their models are the only models, so everything you run yourself lives in a separate window with none of your context.

We built Heuricity around the opposite assumption. A model you bring is a first-class citizen. Connect one and it appears in the same menu as the included catalog, with a hollow marker instead of a solid one. It gets the same conversation, the same tools and memory, the same streaming. The router can pick it when it fits. Every message shows where it ran, so a turn that used your model says so, right on the message.

The local case

Running a model locally used to be a weekend project. Now it is an evening with Ollama or LM Studio and one paste of a URL. What you get is worth being honest about. A small local model is not a frontier model, and pretending otherwise helps nobody. But it is private in a way nothing else can match, it works on a plane, and its marginal cost is electricity. For drafts you would rather keep on your own disk, for quick questions when you are offline, and for the pleasure of watching your own hardware think, it is quietly wonderful.

Why we never meter it

Usage on your own models and accounts never counts against a Heuricity plan. That is not generosity, it is just whose bill it is. Your machine, your electricity. Your API key, your agreement with the provider. What you pay us for is the harness: one place where all of it behaves like one product. The app even shows you what you saved by running things your way.

Where this goes is not hard to see. Models keep getting smaller and better at the same time, and the hardware in your pocket keeps getting stranger. We expect the share of AI that runs on things you own to grow every year. The harness should not care where the model lives. Ours does not.