ask
Send one prompt to an AI backend and retrieve the response. Choose 'auto' for smart routing by task complexity or specify a model to direct the query to a particular provider.
Instructions
Send one prompt to one AI backend and return the response. model:'auto' lets SAB's router pick the best backend by task complexity + current health; passing a specific model name forces that provider. Use this for direct LLM queries that don't fit a more specialized tool. For multi-backend consensus on the same prompt, use council. For agentic multi-step work with a defined role, use spawn_subagent. For LLM-driven file generation or editing, use generate_file / modify_file so the file content stays out of Claude's context window. Read-only: makes one HTTP call to the chosen backend. Returns: {success, model, requested_backend, actual_backend, prompt (truncated preview), response (the LLM output), backend_used, fallback_chain, response_time, cache_status, thinking_enabled, max_tokens, was_truncated, smart_routing_applied, routing, processing_time}.
Input Schema
| Name | Required | Description | Default |
|---|---|---|---|
| model | Yes | AI backend to query: auto (smart routing selects optimal backend), local (autodiscover vLLM/llama.cpp/LM Studio), gemini (Gemini Enhanced, 32K tokens), nvidia_deepseek (NVIDIA DeepSeek with streaming + reasoning, 8K tokens), nvidia_glm (NVIDIA GLM-5.2 code specialist, 32K tokens), openai (OpenAI GPT-5.2, 128K context, premium reasoning), groq (Llama 3.3 70B, ultra-fast 500+ t/s). The friendly aliases `deepseek` and `glm` are also accepted (mapped to nvidia_deepseek / nvidia_glm), matching the other tools. `nvidia_qwen` and `qwen3` are legacy aliases still accepted for back-compat (the lane served Qwen3 Coder 480B until NVIDIA retired it on 2026-06-11) — they resolve to nvidia_glm. | |
| prompt | Yes | Your question or prompt (Unity/complex generations automatically get high token limits) | |
| thinking | No | Enable thinking mode for DeepSeek (shows reasoning) | |
| max_tokens | No | Maximum response length (auto-calculated if not specified: Unity=16K, Complex=8K, Simple=2K) | |
| force_backend | No | Force specific backend (bypasses smart routing) - use backend keys like "local", "gemini", "nvidia_deepseek", "nvidia_glm", "openai", "groq" (legacy "nvidia_qwen"/"qwen3" also accepted) | |
| model_profile | No | Only for the local backend: request a specific model id from your local router (llama-swap/llama.cpp/LM Studio). Valid values are whatever your own router serves — query its /v1/models endpoint for the list; SAB does not ship a fixed set. The id is passed through as the requested model, so an id your router does not serve fails the call (HTTP 400 model not found) rather than falling back. SAB also tries to load the model first if it is not resident, but that pre-flight step only probes http://localhost:8081 — if your server listens on another port the load is skipped (the request still goes to your configured endpoint). Omit to use whichever model is already loaded. | |
| enable_chunking | No | Enable automatic request chunking for extremely large generations (fallback if truncated) |