Local is your next step
Follow the local AI guide to establish a working runtime before connecting the app.
Practical guides · Start here
Choose where your AI runs
Understand the practical differences between a local model, a self-hosted server and an API provider before connecting your app.
Three routes to compare
| Route | Where the model runs | What to plan for |
|---|---|---|
| Local | On your computer, using an installed runtime and model. | Model downloads, disk space, available memory and response speed. Smaller models are often an easier starting point. |
| Self-hosted | On a machine or server you control. | Server availability, network access, access controls and any hardware or hosting costs. |
| Provider API | At the provider you select. | An account and supported key, exact model ID, provider terms, rate limits and possible charges. |
An open-source app and a free website do not guarantee free inference. Provider allowances may change. Your own server also consumes resources even when the app is free to obtain.
Make a choice in five questions
- Where may this information go? If a task includes confidential text, decide whether sending it to a provider or another machine is acceptable before choosing a route.
- What can your computer handle? Start with a model whose download and memory requirements fit comfortably alongside your other apps. A larger model or longer conversation may use more memory.
- How often will you use it? Compare practical local performance with your provider's usage limits and billing. Check the provider account rather than assuming a free profile means unlimited use.
- Does the task need tools? A model that answers chat may not support every tool or capability required by an app. Use the project's compatibility instructions.
- Can you verify the result? Try a small task with a known answer before trusting the route with a larger project.
Connect it deliberately
In TalkToAi Code, read the current setup guide for the model route your installed version supports. In ZERO ONE, configure an existing local runtime, OpenZero server or supported Assistant API in Settings. Match the endpoint and model ID to the service you actually have.
When enabling alternatives or fallback profiles, consider each destination separately: a switch can send task context to that alternative provider. Check the app's visible route information and provider usage account when troubleshooting.
An API is your next step
Use the provider's own account and documentation, then the app's current README. Keep keys out of prompts, screenshots, public issues and source control.