llmlisted

# LLM Calls `bin/llm` to query an external LLM gateway. ## When to use - User asks to consult another model ("what does GPT think", "get a second opinion", "ask Gemini") - User wants to compare or debate responses across models - On your own initiative: cross-check a non-obvious claim, validate a design decision, or get a second perspective before committing to a recommendation ## How to call ```bash bin/llm [options] "message" echo "message" | bin/llm [options] ``` Options: - `-m MODEL` — model name (default: `gpt-5.6-sol`) - `-s "SYSTEM"` — system prompt - `-t N` — max output tokens (default: 32768) - `--temperature F` — sampling temperature (suppressed when `--thinking` is set) - `--thinking` — enable reasoning before final response; intermediate thought is hidden - `--effort low|medium|high|xhigh` — reasoning effort when `--thinking` is set (default: `high`) - `--stream` — stream output progressively **Reasoning effort per model family:** | Family | Parameter | Notes | | ------ | --------- | ----- | | `Claude-*-4.6` and older | `budget_tokens` | low=1024 medium=4096 high=16000 xhigh=32768; clamped if near `--max-tokens` | | `Claude-*-4.7+` | `output_config.effort` | adaptive mode via gateway | | `o1/o3/o4-*` | `reasoning_effort` | always on; xhigh→high | | `gpt-5.*` | `reasoning_effort` | opt-in reasoning; all levels supported | | `DeepSeek-*` | `reasoning_effort` | low/medium/high→high, xhigh→max | | `gemini-*` | `thinkingBudget` | low=512 medium=4096 high=16384