uncommit the dead

Every ghost needs
a local medium.

Install Ollama. Download a model. Bring your code back.

Download README

Install Ollama

Ollama runs downloaded models on your computer. For this app, use a local open-weight model. No API key or cloud account is needed.

Download for macOS, Windows, or Linux

Follow the installer’s instructions. On macOS or Windows, open the Ollama app. In Terminal or PowerShell, check the installation:

ollama --version
Using Linux or a terminal-only setup?

If the server is not already running, start it and leave that terminal open:

ollama serve

“Address already in use” can mean Ollama is already running. Check the connection before starting another instance.

Choose your medium.

Start with Gemma 3 4B. Run this from any folder and wait for the download to finish:

ollama pull gemma3:4b
Local Gemma 3 options
ModelDownload*When to use it
gemma3:1b806 MBQuick setup check; plans can be generic.
gemma3:4b3.4 GBSuggested starting point for this app.
gemma3:12b8.2 GBOptional, if your hardware has room.

*Approximate listed sizes, not RAM requirements. Runtime memory and speed depend on hardware and context. Check the official model library for current tags and licences.

Need a smaller download?
ollama pull gemma3:1b

The 1B model is useful for checking the setup, but may struggle with structured plans and evidence.

Check your installed names, then try a prompt:

ollama ls
ollama run gemma3:4b "Suggest one tiny feature for an unfinished todo app."

The first request may take longer while the model loads. For interactive chat, run the model without a prompt; use /bye to leave.

Connect to the graveyard.

  1. Open Graveyard, then Connect local AI or Local AI connected.
  2. Click Check connection after downloading a model.
  3. Select the exact installed name, such as gemma3:4b.
  4. Add a repository with the folder picker, or choose a sample.
  5. Choose an afterlife and click Summon its last wish.

Review the evidence, then download the plan. Suggestions are not executed, and the app does not modify your repository.

Keep it local: use downloaded model tags, rather than cloud variants such as :cloud. This app connects to Ollama at 127.0.0.1:11434 on the same machine.

After the app dependencies and model weights are installed, local inference can work without internet. Other local chat models supporting structured JSON can also be selected; embedding-only models cannot write a plan.

Try your first séance

Keep the attic tidy.

See downloaded models and those currently using memory:

ollama ls
ollama ps

Unload a model from memory while keeping its download:

ollama stop gemma3:4b
Remove a downloaded model

This removes its local installation. You will need to pull it again to use it later.

ollama rm gemma3:4b

When the séance goes quiet.

Ollama is offline or the connection is refused

Open Ollama or run ollama serve, then check the connection again. Open the local model-list endpoint in your browser: you should see JSON with a models array. Remote endpoints are not configurable in this version.

Connected, but no models appear

Finish a model download, confirm it with ollama ls, then click Check connection. An empty model list means no models are installed in that Ollama instance.

The model is unavailable

Use the exact name shown by ollama ls, including its size tag. Refresh the app’s model list after downloading or removing models.

Generation is slow or times out

The app allows up to two minutes per request. Try a smaller model, close memory-heavy apps, or unload an unused model with ollama stop MODEL_NAME. Smaller models may be faster but produce less useful plans.

The plan is incomplete or has no valid evidence

Retry once or choose a more capable local instruction model. Responses must contain structured fields and references matching the repository. A clear README and useful TODOs help; every model is not equally reliable.

The repository will not import

Choose a Git repository with at least one commit and check folder permissions. Plain source folders are not Git repositories. Selecting a subfolder inside a repository resolves its root.

The terminal cannot find the ollama command

Reopen Terminal or PowerShell after installation. If it still fails, follow the installer’s platform-specific instructions and check your PATH.

Checked against official documentation on October 28, 2026.

Quickstart ↗CLI reference ↗Model-list API ↗