Skip to content

Overview

auto/lab

git-for-state, slurm-for-compute. Drive autonomous research from your terminal — create a project, queue experiments, and attach GPUs as execution nodes. The agent writes code, runs jobs, and analyzes results; you stay in the loop.

Quickstart Install & setup

curl -fsSL https://autolab.ai/install.sh | sh   # installs the `autolab` CLI
autolab login                                   # sign in (opens a browser)
autolab init                                     # turn this repo into a project
autolab start                                    # go live — the agent starts researching

What is Autolab?

Autolab runs experiments for you. An LLM agent drives the research loop: it proposes hypotheses, writes experiment code, schedules jobs across your hardware, analyzes the results, and merges the improvements — autonomously. You stay in the loop through the CLI and the dashboard, adding ideas, setting the objective, or steering the queue whenever you want.

The autolab CLI is the terminal gateway to all of it. It treats your workspace like git: a project is your objective plus a baseline, an experiment is a commit, and submit queues one. No server stack required — it's a slim client that wraps the control node's API.

The mental model

   you ── autolab CLI ──▶  Control Node (app.autolab.ai)  ──▶  Execution Nodes
                          the agent + queue + dashboard          your GPUs / machines
  • Control Node — hosted at app.autolab.ai. Home of the agent, the experiment queue, and the dashboard. It also serves each project's private git repo.
  • Execution Nodes — any machine you attach with autolab serve. They clone the project's code and run the experiments. Heterogeneous by design: connect a laptop, a workstation, or an 8×H100 box.
  • The CLI — autolab on your machine. One binary to create projects, submit experiments, manage settings, and turn machines into execution nodes.

See Core concepts for how projects, experiments, and keys fit together.

Start here

  •  Quickstart

    From install to your first running experiment in five minutes.

    Quickstart

  •  Core concepts

    Projects vs. experiments, the commit == experiment model, and field locking.

    Concepts

  •  Run experiments

    submit with code, as an idea, or as a seed — plus diff, checkout, and logs.

    Run experiments

  •  Execution nodes

    Turn any machine — including a headless GPU box — into a runner with serve.

    Execution nodes

  •  CLI reference

    Every command, argument, and option — generated from the CLI itself.

    CLI reference

  •  Configuration

    Hosts, profiles, environment variables, and the .autolab/ workspace.

    Configuration