~/launch80 $[discord]Join the community - talk local AI with other engineers
Documentation menu

Rent a GPU

L80 vast runs bbvastl80: rent a Vast.ai GPU, serve a model on it with vLLM or llama.cpp, run BetterBench from your machine against it, publish, and destroy the box. Every bbvastl80 command works through it.

Free, no token

Commands

$ L80 vast doctor                              # Vast key, SSH key, and the L80 and BetterBench it will call$ L80 vast search --spec run.yaml              # matching offers and prices; does not bill$ L80 vast run --spec run.yaml --max-hours 2   # rent, serve, benchmark, publish, tear down$ L80 vast status --json                       # every run, and which may still be billing$ L80 vast destroy --run <id>                  # asks before destroying

L80 points bbvastl80 at this L80 for publishing and at the pinned BetterBench for measuring, so a rented run produces the same page as a local one. Publishing still needs a token.

Nothing left billing

  • --max-hours N is a watchdog. When time runs out, L80 stops the run and destroys its instance.
  • After any run that fails or is interrupted, L80 names each instance that may still be billing and the command that stops it.
  • The first Ctrl-C goes to the tool so it can clean up. A second one forces it to stop.
  • L80 never answers a confirmation prompt for you.

Keys and files

  • The Vast key comes from VAST_API_KEY or ~/.config/bbvastl80/credentials.
  • Runs are recorded in ~/.local/state/L80/vast/runs; BBVASTL80_RUNS_DIR moves them. status --json reads these files.
  • L80_VAST runs your own bbvastl80 instead of the pinned one.
  • Agents: L80 skills link --target claude-code --name l80-vast.