Quickstart¶

Install¶

Just the toolingFull agent (revive + Scout)

pip install lazarus-bio

Gives you the commit-era dependency pinner and the contract/compose tooling. Pure Python, no Docker.

pip install "lazarus-bio[agent]"   # needs Python >= 3.10 + Docker

Adds the autonomous resurrection loop and the URL Scout, which drive Claude via the Claude Agent SDK.

Prefer zero install? Open the notebook in Colab — a 2-minute tour with no Docker or GPU.

1. Pin dependencies to a repo's commit era¶

The single biggest reason old code "won't install" is that pip gives you today's versions. Lazarus reconstructs the dependency universe as it was on the repo's last commit — no repo execution required.

lazarus pin --date 2019-01-01 tensorflow numpy scipy
#   tensorflow==1.12.0   (matches MaSIF's real Dockerfile, not its README's 1.9)

2. Resurrect a repo — from just a URL¶

The Scout reads the repo + paper, writes the goal and a falsifiable sanity check, picks a base image, and pauses for your OK before spending compute:

lazarus resurrect https://github.com/jertubiana/ScanNet

Or drive it by hand with an explicit image + goal (both override the Scout):

lazarus resurrect --image pablogainza/masif:latest --workdir /masif \
  --goal-file examples/masif_site_goal.txt --keep

Where it executes

Lazarus runs on your machine; where it executes is one flag. Point --docker-host at a local daemon, a remote x86 box (ssh://you@host), or a cloud/GPU rental — for methods whose binaries need hardware a laptop can't emulate.

3. Compose revived bricks into a pipeline¶

lazarus run examples/pipelines/binder_triage.yaml \
  --input structure=4ZQK.pdb \
  --registry examples --registry components \
  --docker-host ssh://you@your-x86-gpu-box

See Compose & reproduce for what the pipeline concluded on PD-L1.

Auth¶

Log in the claude CLI (subscription) or drop ANTHROPIC_API_KEY=... in a gitignored .env — lazarus loads it without ever putting the secret on the command line.