Running an AI Chatbot Without the API Bill

Cloud AI APIs are the default for a reason: they're easy. But easy comes with three recurring costs — the monthly bill, the data leaving your control, and the lock-in to whichever vendor you built on. For sensitive or high-volume chatbot use cases, those costs compound.

The build

The private alternative is smaller than people expect. A webhook receives incoming chat messages. n8n structures the conversation history and sends it over HTTP to an Ollama instance — an open-source runtime for local AI models — running on a machine in the office. The response gets parsed and returned as a clean reply. Tailscale provides secure remote access without exposing a single port to the internet.

Every conversation stays on hardware you own. There is no per-message fee, no usage tier, and no terms-of-service change that can break your product.

The gotcha

Local models are not cloud frontier models. For open-ended reasoning, the big hosted models still win. But most business chatbots don't need frontier reasoning — they need to answer questions from a known set of information, consistently, privately, and cheaply. Matching the model to the actual job is the whole game.

The result

API costs eliminated — conservatively $100–$400 a month at moderate chatbot volume, and it scales with usage — with full control over the model and the infrastructure, and conversation data that never leaves the local network.