Using AfterImage with Local Models

AfterImage supports any OpenAI-compatible local model server. No API key needed.

Quick start

pip install afterimage
afterimage generate -c examples/configs/local.yaml

vLLM

pip install vllm
vllm serve Qwen/Qwen3-1.7B --port 8000

Config:

model:
  provider: local
  base_url: http://localhost:8000/v1
  model_name: Qwen/Qwen3-1.7B

Ollama

ollama pull llama3.2
ollama serve

Config:

model:
  provider: local
  base_url: http://localhost:11434/v1
  model_name: llama3.2

llama.cpp

./llama-server -m model.gguf --port 8000

Config:

model:
  provider: local
  base_url: http://localhost:8000/v1
  model_name: model

Tips

• Lower max_concurrency for CPU inference (1-2 is usually best)

• max_turns: 1 keeps generation fast for small models

• Quality gating (auto_improve: true) requires local embeddings: pip install "afterimage[embeddings-local]"