Ollama (local models)

Ollama is a free app that runs open models on your computer. When a chat node uses a local model, its conversation and connected context go to Ollama on your machine instead of a cloud provider. Long research sessions cost nothing per message, conversations stay on your computer, and chats work offline. The trade-off: your hardware sets the speed and the quality ceiling.

Set up Ollama

1. Open Settings and click Ollama under AI Settings. If Slashspace can't reach Ollama, the page shows Please Start / Install Ollama.
2. Click Install Ollama to open Ollama's download page. Install Ollama and start it (open the app, or run ollama serve).
3. Leave the Ollama settings page and open it again. Once Slashspace finds Ollama, the page shows Installed and Ollama's version.

Download a model

Models you've downloaded into Ollama appear in Slashspace automatically. To get one, type its name (for example llama3.2) into the field on the Ollama settings page and click Download Model. Search Models on Ollama opens Ollama's model library. You can also run ollama pull llama3.2 in a terminal.

Models loaded in memory show Running. Delete removes a model from your computer.

Choose a model whose download size sits well under your computer's memory (or your graphics card's). Small models (a few billion parameters) suit most laptops; bigger ones answer better but slower.

Use a local model

Open the model picker in a chat node and choose a model from the Ollama group. To start new chats on it, make it your default model.

• Cost: local models don't draw from the hosted wallet. Like every chat, they need an active plan: the trial, the subscription or Lifetime. See Plans and billing.
• Agent mode: new chats start in Agent mode, which gives the model tools, so only models that support tool calling work there. For other models, switch the chat to Chat mode (⌘2, or Ctrl+2 on Windows and Linux). Ollama's library tags models that support tools.
• Images: for connected Image nodes or a Scratch pad, pick a model tagged vision.
• Orchestrator and Image modes don't use Ollama.

What still uses the network

• Sign-in: Slashspace checks your sign-in before each chat and renews it while you're online. After a couple of days offline, local chats can stop with a sign-in error.
• Capturing sources: Web and Post capture and Document processing run on hosted services. Once captured, the text is saved with your canvas.
• Chat titles: by default, a hosted service names each new chat from its first message.
• Connectors reach their own services, and voice input transcribes in the cloud.
• Usage analytics (not your messages) go to PostHog. See Privacy and your data.

Connection settings

The gear next to Ollama's version (or Configure when Ollama isn't found) opens Ollama Configuration. Hostname is where Slashspace looks for Ollama, http://localhost:11434 by default; use the default restores it.

Note: Chats always go to Ollama on this computer at the default address, whatever the hostname says. Keep Ollama running locally on port 11434.

Troubleshooting

• Ollama isn't detected. Open http://localhost:11434 in a browser; it should say "Ollama is running". Check that Hostname is the default, then reopen the settings page.
• Models are missing from the picker. If you started Ollama after Slashspace, they can take up to five minutes to appear. Restarting Slashspace shows them right away.
• Download Model does nothing. An unknown name fails silently. Copy the exact name from Ollama's library, or run ollama pull in a terminal to see the error.
• Replies are slow. The first reply after a pause waits while Ollama loads the model. Try a smaller model and connect only the nodes a chat needs.
• Answers miss part of a long context. Ollama cuts off anything beyond its context length, which you can raise in Ollama's settings.

More fixes in Troubleshooting and support.