Connecting an AI provider
Lucius needs a model to talk to. Connect one during first run or later from Settings → AI Model — same screen either way; first run just shows the short version.
The two ways to connect
Existing CLI login. If the Claude Code CLI or Codex CLI is already signed in on this machine, the app detects it and offers a one-click “Use this.” No API key stored — Lucius spends the model access your existing Claude Pro/Max or ChatGPT subscription already gives you.
Paste an API key. Pick a provider and enter its key. It’s sent to your own instance, stored server-side, and never returned to the browser or shown again — Settings only shows whether a key is set, never the key itself.
Supported providers
- Anthropic — Claude models directly (needs an API key).
- OpenAI — GPT models directly (needs an API key).
- Ollama — a model running locally or on your own network (no key; point it at your Ollama server).
- A full catalog in Settings also covers Gemini, xAI, Zhipu (GLM), OpenRouter, LiteLLM proxies, and any other OpenAI-compatible endpoint — first run only shows the short list above; everything else is one click away in Settings once you’re in.
What it costs
Open Net Worth charges nothing — you pay your model provider directly, same as using their API from any other tool. Cost depends on which model you pick and how much you use Lucius; there’s no single honest number to quote.
What the app gives you instead is visibility and a ceiling:
- Settings → AI Model → Cost shows real spend — today, yesterday, the last 7/30/90 days, and all-time — broken down by model and by day, from actual token usage.
- A monthly budget cap, with a progress bar, so a runaway conversation doesn’t surprise you on the bill.
- Cost controls under the hood: a memory toggle, cheaper models for background/automation tasks, conversation folding (old context isn’t re-sent every turn), and a tripwire for a tool called on repeat — all keeping spend proportional to what you’re actually asking for.
Skipping this entirely
Run Open Net Worth with no AI provider at all. Accounts, the ledger, budgeting, investments, and CSV/OFX/QIF import all work without one. What you lose: chat with Lucius, reading PDF statements, auto-categorization, automations, and cross-session memory — anything that needs a model to reason about your data.