llmwalk explores the answer-space of prompts using MLX-supported open LLMs by branching completions based on top-k, top-p, and temperature parameters.
It prioritizes likely branches in a tree search until finding the n most probable finished answers, stopping early without full enumeration. Usage involves uvx llmwalk with options like -p for prompt, -m for model, -n for answer count, and --format for JSON/CSV output.
JetBrains’ reasoning-augmented code MoE (12B total / 2.5B active) that emits explicit <think> chains for debugging, planning, and agentic coding
