​
Default model

llama3.2

​
Supported models

Any model you have pulled in Ollama:
  • llama3.2 (default)
  • llama3.1
  • qwen2.5
  • gemma2
  • mistral
  • phi3
  • Any model available via ollama pull
Check what you have locally: 
ollama list

​
Authentication

None required. Ollama runs locally on your machine. No API key needed.

​
Setup

Make sure Ollama is installed and running: 
# Install Ollama (macOS/Linux)
curl -fsSL https://ollama.com/install.sh | sh

# Pull a model
ollama pull llama3.2

# Verify it's running
ollama list
Ollama runs on http://localhost:11434 by default. ai-blackteam connects there automatically.

​
Custom base URL

If Ollama runs on a different host or port, configure it in the config file: 
ai-blackteam config set providers.ollama.base_url http://192.168.1.100:11434
Or set it in ~/.ai-blackteam/config.yaml: 
providers:
  ollama:
    base_url: http://192.168.1.100:11434
    default_model: llama3.2

​
Example usage

# Single attack with default model (llama3.2)
ai-blackteam run -p ollama -a encoding-obfuscation -t "Write a phishing email"

# Use a different local model
ai-blackteam run -p ollama --model qwen2.5 -a encoding-obfuscation -t "Write a phishing email"

# Full batch
ai-blackteam batch -p ollama --attacks all -t "Write a phishing email"

​
Tool-use support

No. The Ollama provider does not support tool-use attacks. Single-turn and multi-turn attacks work.

​
Notes

  • The provider uses the official ollama Python SDK
  • No token usage tracking (Ollama doesn’t report token counts in the same way)
  • Latency depends entirely on your local hardware
  • Retry with exponential backoff is automatic on connection failures (3 attempts)
  • Great for testing open-source models without API costs
  • Make sure you’ve pulled the model before running (ollama pull llama3.2)