Skip to content

OASIS ScienceClaw

ESIIL's multi-agent workspace

Run an environmental synthesis workspace on your laptop, keep agent access narrow, and publish reviewed outputs without mixing private workspace files into the public site.

Start here First 10 minutes What is a container? Launch locally Template launch Where files go Troubleshooting

  • Beginner-safe path

Learn the mental model first, then launch the container and do one small useful action.

  • Portable scientific workspace

Think of the container as a field station in a box: a reproducible set of tools for shared work.

  • Repo as memory

Keep decisions, assumptions, prompts, tasks, reports, and template improvements in versioned project memory.

  • Narrow, safer mounts

Mount only the folders agents need. Keep secrets local and large data in external storage.

  • Scientific working group

Use an 11-role environmental data science scaffold with a PI Liaison, shared memory, evidence standards, skeptic review, and human approval gates.

  • Visible outputs

Find figures, reports, logs, notebooks, maps, and provenance in predictable workspace folders.

  • Recoverable by design

Treat troubleshooting as normal work: checkpoint, inspect, recover, and avoid destructive commands.

  • Advanced when ready

Add model routing, storage backends, GitHub workflows, branded skins, and Kubernetes workers after the basics feel stable.

Quick Win

cp .env.example .env
make init-working-group
make doctor
docker compose build
scripts/login-codex.sh
make checkpoint

The First 10 Minutes page gives the calm walkthrough. The Glossary explains container terms without assuming command-line fluency.

Read Security before connecting Slack tokens to the PI Liaison.

For deeper work, continue to Template GitHub Launch, OASIS ScienceClaw Template, Model Routing, Storage, Publishing Workflow, and Distributed Runtime.