⚡ CEM

   ██████╗███████╗███╗   ███╗
  ██╔════╝██╔════╝████╗ ████║
  ██║     █████╗  ██╔████╔██║
  ██║     ██╔══╝  ██║╚██╔╝██║
  ╚██████╗███████╗██║  ╚═╝ ██║
   ╚═════╝╚══════╝╚═╝      ╚═╝

One command, many AIs. One AI thinks, another writes the code.

Use a strong model for the thinking and a cheap one for the typing — you get better decisions without paying premium rates for every line of code.

curl -fsSL cem.pw/install | sh        # macOS / Linux / WSL

irm cem.pw/install | iex              # Windows (PowerShell)

JetBrains IDE plugin: Settings → Plugins → Marketplace → search cem (listing). VS Code and the other editors: README_DETAILS.md.

cem "what is a B-tree?"           # ask the thinker
cem -w "write a quicksort in Go"  # ask the writer
cem -p "add retries to client.go" # pair: thinker plans, writer codes

Local or self-hosted models work the same way — ollama, LM Studio and unsloth are tools too, you just give cem the address:

cem endpoint ollama 192.168.1.10:11434 --model qwen3-coder
cem roles ollama claude            # ollama thinks, claude writes

The first run opens a short wizard: pick your language, pick which AI thinks and which one writes. That's it.

  🧠 THINKING · gpt  gpt-5.6-terra · xhigh
  - File: retry.go
  - Function: func withRetry(fn func() error, n int) error
  - Edge cases: context cancellation, exponential backoff cap
  ⏱ thinking 8.9s
  ────────────────────────────────────────────────────
  ✍️  WRITING · claude  sonnet · low
  retry.go created — withRetry retries n times with capped backoff.
  ⏱ writing 24.1s
  ⏱ total 33.0s   (thinking 8.9s + writing 24.1s)

• README_DETAILS.md — every command, IDE plugins, API keys, models, reasoning effort, troubleshooting
• ADVANCED.md — project configs, build from source, internals
• Türkçe: README.tr.md
• Site: cem.pw

MIT — see LICENSE.