[Sym-jay]

clishe setup (also setup inside a session) gets someone from zero to a working local model without reading docs:

$ clishe setup
Clishe: Checking this computer for a local AI model...

  Memory: 16 GB
  A model that fits: qwen2.5-coder:7b (7B, good with shell commands, about 5 GB)

  ✓ Ollama is running. Models downloaded: llama3.2:latest
    Get the suggested model with: ollama pull qwen2.5-coder:7b

  ✓ Local only: nothing you type is sent to the internet.

Download qwen2.5-coder:7b now? This runs: ollama pull qwen2.5-coder:7b [y/N]:
Use qwen2.5-coder:7b for Clishe? [Y/n]:

• Memory → model: under 8 GB gets llama3.2:1b, 8–16 GB gets llama3.2 (3B), 16 GB and up gets qwen2.5-coder:7b.
• Ollama: says whether it's not installed (with the download link), installed but not running (ollama serve), or running (lists its models).
• Other servers: detects llama.cpp, LM Studio, Jan and vLLM servers on their usual ports.
• Asks before doing anything: downloading is [y/N]; saving the model to the config is [Y/n]. The config keeps its other settings and 0600 permissions.
• Warns if the cloud provider is on, since that's the only way anything could leave the machine.

Tests

• 15 new pytest tests: memory thresholds, Ollama's :latest tag, report wording for each state, config save. 466 pass locally.
• Manual run against a fake Ollama on port 11434: the report, the "use it?" question, then the config was saved and a second run said "All set".

🤖 Generated with Claude Code