TERM–LLM / LEARNA REAL TERMINAL. THIRTEEN SMALL STEPS.

FROM YOUR FIRST COMMAND TO A CONVERSATION

Learn by doing.

Your terminal on the left. A little guidance on the right.
Learn term-llm without installing anything.

  1. 01 · Explore the shell
  2. 02 · Ask, act, connect
  3. 03 · Have a conversation
Session settings Qwen3 8B · 16K

Qwen mode runs on your GPU, not a cloud inference service. Simulator mode uses scripted responses, not a model. It needs roughly 7–10 GB of available GPU memory (more context uses more memory). Linux uses an additional 512 MB. Answers can still be wrong—check commands and results.

Reloading or shutting down discards guest files and conversations. The checklist lesson lets you save your work. Nothing runs on your host filesystem.

Terminal your commands · your workspace

Type commands in the terminal. Back / Next never run commands or reset your files. In chat, /quit returns to the shell.

Behind the scenes

Real Linux and the native term-llm CLI in both modes. Qwen3 8B runs through WebGPU; Simulator is a clearly labelled scripted provider for practising the same lessons. Tool requests and file contents are genuine. Lesson navigation never fabricates model output or marks model prose as correct. The local picnic MCP server has no network access.

Runtime allocation
Last model output — may contain your input
Raw Linux serial