Local Models

Run Victor with local LLMs for privacy, offline use, and predictable cost.

Provider Comparison

Provider	Setup	Air-Gapped	Features	Best For
Ollama	brew install ollama	✅	Tool calling, streaming	Quick start
LM Studio	Download app	✅	GUI model management	Desktop users
vLLM	pip install vllm	✅	High throughput	Production
llama.cpp	Compile from source	✅	CPU inference	No GPU

Quick Start

# Option 1: Ollama (recommended)
victor init --local

# Option 2: Manual setup
ollama serve && ollama pull llama3
victor chat --provider ollama --model llama3

Configuration

# ~/.victor/config.yaml
profiles:
  local:
    provider: ollama
    model: llama3

providers:
  ollama:
    base_url: http://localhost:11434

Recommended Models

Provider	Model	Size	Use Case
Ollama	llama3	8B	General coding
Ollama	codellama	13B	Code completion
Ollama	mistral	7B	Fast responses
vLLM	mixtral	8x7B	Best quality

Air-Gapped Mode

# Disable all network tools
victor init --airgapped

# Or enable per-session
victor chat --airgapped

Commands Reference

Provider	Start Server	Pull Model
Ollama	ollama serve	ollama pull <model>
LM Studio	Start in app	Download via GUI
vLLM	python -m vllm.entrypoints.openai.api_server --model <id>	Manual