Local AI setup

Ollama for Actium

Actium's AI-assisted features run through a local Ollama server on your Mac. That keeps semantic scoring, catalogue generation, and right-click AI processing close to your own files.

ollama pull nomic-embed-text:latest
ollama pull qwen2.5:7b-instruct
ollama pull gemma4:e2b

Setup overview

Install Ollama once, then pull the models Actium needs.

01

Install Ollama

On macOS 14 Sonoma or later, download Ollama from the official site, open the app, and allow it to make the ollama command available in Terminal when prompted.

Download Ollama for Mac
02

Check the local server

Actium expects Ollama at http://localhost:11434 by default. Leave the Ollama app running before using local AI features.

ollama list
03

Pull a model

Use ollama pull for each model you want Actium to use. Pulling downloads the model locally, so allow disk space and time.

ollama pull model-name

Terminal install option

Ollama also publishes a terminal installer:

curl -fsSL https://ollama.com/install.sh | sh

Recommended minimum models

Models shown in the current Actium settings.

Actium feature What it does Recommended minimum model Pull command
Auto-file Embedding scoring and Related Items Builds local embeddings so Actium can score document similarity by meaning rather than word overlap. nomic-embed-text:latest ollama pull nomic-embed-text:latest
Semantic Catalogue Generates local concept summaries and relationships for a selected item or repository context. qwen2.5:7b-instruct ollama pull qwen2.5:7b-instruct
Ollama Processing Runs right-click "Process with AI" requests using a prompt you write, locally on this Mac or queued for another registered Mac. gemma4:e2b ollama pull gemma4:e2b

These model names reflect the current Actium settings screenshots. Cosine and Bayes Auto-file scoring do not require Ollama. Embedding scoring, Semantic Catalogue, and Ollama Processing do.

How Actium uses Ollama

Three local-AI paths, each with a different model need.

Embeddings

Semantic file similarity

Actium can create embeddings in the background as files change, improving Auto-file suggestions and Related Items over time.

Semantic Catalogue

Concept graphs

Right-click an item and ask Actium to summarize key concepts and relationships as a graph, with the document text staying on the Mac.

Ollama Processing

Prompt-driven processing

Run a file through a local model with your own prompt. If this Mac is request-only, another registered Mac can pick up the work.

Actium settings

Use the same server address and model names in Actium.

In Actium Desktop Settings, the screenshots show the Ollama server set to http://localhost:11434. Once models are pulled, use Refresh Models in Actium and choose the model for each feature.

# Embeddings
nomic-embed-text:latest

# Semantic Catalogue
qwen2.5:7b-instruct

# Process with AI
gemma4:e2b