TALKTOAI

Practical guides · Start here

Start with a working local model

A practical connection guide for people who want a model on their own computer, with clear checks for the runtime and desktop app.

Understand the two pieces

The desktop app organises your work. A runtime loads the model and answers requests. Installing TalkToAi Code or ZERO ONE does not by itself establish a working local model service.

Ollama is one runtime option. Install it from Ollama's official download page. Follow the official quickstart to select and run a local model. Choose a download that fits your available disk space and memory; begin with a smaller model and a short task.

Connect one step at a time

  1. Get a local answer first. Open the runtime and select a local model. Ask it to write a short sentence. Resolve runtime errors before changing app settings.
  2. Check the service address. Ollama's documented local API uses http://localhost:11434. This refers to the computer running the client. A service on a different machine needs that machine's configured address.
  3. Follow your installed app's setup guide. Use TalkToAi Code's README or ZERO ONE's README for supported routes, endpoint format and model selection. Exact settings depend on the release.
  4. Try a harmless prompt in the app. Ask for a brief explanation of a familiar topic. Check the selected route and model before adding project files.
  5. Increase the task gradually. Start with one file or a short note. Longer context and larger models can increase memory use and slow responses.

On ZERO ONE's Microsoft Store edition, configure an existing runtime or API. The product page states that Store editions do not install local models. Follow the repository's separate instructions for direct editions.

If the connection fails

Connection refused or service unavailable

Confirm the runtime is running on the expected machine and address. A local service must be available before an app can use it. On Linux, Ollama's quickstart documents ollama serve when the server is not already running.

Model not found

Check the model installed in the runtime and use its exact name in the app. Selecting a model in an app is not proof that its files have been downloaded.

It works in the runtime, but not the app

Compare endpoint format and route type with the current app instructions. Record the error, installed app version and runtime version for support.

It responds slowly or runs out of memory

Try a smaller model, a shorter prompt and fewer open applications. Ollama documents ollama ps to inspect loaded models and their CPU/GPU allocation in its FAQ.

Local inference and network activity

A local endpoint tells you where that request is sent. It does not prove the whole workflow is offline. Model downloads, optional APIs, cloud models, web tools, mail connectors and remote servers can use the network. Review the enabled route and integrations before using sensitive material.

Keep a local model service bound to its intended interface. Publishing a model endpoint on the internet is a separate administration task; do not change network exposure simply to fix a desktop connection.

Next: collect a useful, redacted bug report if the runtime works but the app still fails.