Ollama¶

QUASAR uses cloud models via Ollama for fast execution and best performance.

Installation¶

macOS¶

brew install ollama

Linux¶

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

Windows¶

Download from ollama.ai/download

Start Ollama¶

ollama serve

Default URL: http://localhost:11434

Pull Cloud Models¶

QUASAR uses cloud models by default. Pull them for Auto mode:

# Required for Auto mode
ollama pull glm-4.7:cloud
ollama pull deepseek-v3.1:671b-cloud
ollama pull qwen3-coder:480b-cloud
ollama pull gpt-oss:120b-cloud

Available Models¶

Cloud Models (Default)¶

Model	Description	Tool Calling
`glm-4.7:cloud`	Primary, balanced	✅ Excellent
`deepseek-v3.1:671b-cloud`	Code intelligence	✅ Excellent
`gpt-oss:120b-cloud`	Large, powerful	✅ Excellent
`qwen3-coder:480b-cloud`	Coding focused	✅ Good
`deepseek-v3.2:cloud`	Latest DeepSeek	✅ Good

Recommended

glm-4.7:cloud is the default model in Auto mode.

Usage¶

# Auto mode (uses cloud models)
quasar "your request"

# Specify a cloud model
quasar --model ollama/glm-4.7:cloud "your request"
quasar --model ollama/deepseek-v3.1:671b-cloud "analyze this code"

Custom/Local Models¶

QUASAR also supports any local model for flexibility:

# Use custom local models
quasar --model ollama/qwen2.5-coder:7b "your request"
quasar --model ollama/codellama:7b "explain main.py"
quasar --model ollama/deepseek-coder:6.7b "refactor utils.py"

When to Use Local Models¶

Cloud model rate limits exceeded
Offline work required
Privacy-sensitive tasks
Experimenting with different models

Pull Local Models¶

ollama pull qwen2.5-coder:7b
ollama pull codellama:7b
ollama pull deepseek-coder:6.7b

Configuration¶

No API key needed! Ollama is detected automatically.

To specify a custom URL:

OLLAMA_BASE_URL=http://localhost:11434

Troubleshooting¶

Connection Refused¶

Error: Connection refused to http://localhost:11434

Solution: Start Ollama:

ollama serve

Model Not Found¶

Error: Model 'xyz' not found

Solution: Pull the model first:

ollama pull glm-4.7:cloud