LevelUpWith

Explore / Science

Claude Science: A Researcher's Working Guide

Illustrated cover for the course “Claude Science: A Researcher's Working Guide”

Run reproducible analyses in Anthropic's beta desktop research workbench, from a first sandboxed Python or R run to remote compute beside your data. You will learn to trace every figure to its code and environment, use the reviewer within its limits, connect data sources from any field, and keep sen

Capable · 18 levels · 2 free · Created Oct 2026 · Professionally curated by levelupwith.com

What's inside

  1. Level 1: What Claude Science Is and Is NotFree
    Learn what Claude Science is (a beta desktop research workbench from Anthropic for macOS, Windows and Linux on paid plans), that it is not for clinical or diagnostic use, and that this course is independent of Anthropic and must be checked against Anthropic's current pages.
  2. Level 2: Installing and Running a First AnalysisFree
    Install the app on your operating system, complete sign-in and the setup wizard, and run one small analysis from prompt to saved result.
  3. Level 3: Projects, Sessions, Workspace and Data Directory
    Understand how work is organised into projects and sessions, where the per-session workspace sits, and what the local data directory holds.
  4. Level 4: Data Flow: What Stays Local, What Reaches Anthropic
    Trace what remains on your computer and what is sent to Anthropic as part of the conversation when Claude reads a file, runs code or calls a connector.
  5. Level 5: Permission Cards, Scopes and Revoking Grants
    Read permission cards accurately, choose the narrowest workable scope for each approval, and review and revoke standing grants.
  6. Level 6: The Sandbox, Network Allowlist and Plans
    Understand how the sandbox confines Claude's code, how the deny-by-default network allowlist works, and how to steer multi-step work through an approved plan.
  7. Level 7: Kernels, Cells and the Notebook
    Learn how code executes in persistent Python and R kernels as cells, and how the session's Notebook records each cell, its output and its environment.
  8. Level 8: Environments, Package Installs and GPUs
    Work with the read-only starter environments, create and reuse named task environments, install packages so they persist, and understand local GPU access.
  9. Level 9: Artifacts, Versions and Comments
    Manage saved outputs as artifacts, compare immutable versions, and direct revisions by pinning comments to parts of an artifact.
  10. Level 10: Provenance: Tracing a Figure to Its Code
    Use an artifact version's Provenance record to trace a result back through its messages, code, execution log and environment.
  11. Level 11: The Reviewer: Checks, Limits and Findings
    Understand what the reviewer checks against the execution record, what it does not do, how to work through findings, and how to add your own criteria and control auto-review.
  12. Level 12: Specialists, Delegation and Memory
    Shape how a session works by choosing or customising a specialist, splitting work into parallel tracks with delegation, and controlling what Claude remembers across sessions.
  13. Level 13: Connectors and Their Licence Terms
    Use Featured, Directory and custom connectors to reach outside data sources, and check each source's licence terms before relying on its data.
  14. Level 14: Skills and Literature Access
    Extend Claude's methods with featured, self-written and GitHub-imported skills, and configure literature retrieval through open access, publisher keys or a library proxy.
  15. Level 15: Remote Compute, Cloud Storage and Credentials
    Choose where heavy work runs: an SSH host or Slurm cluster, your own cloud provider account, a scientific model endpoint, cloud storage, or the whole app on a remote Linux server beside the data.
  16. Level 16: Safeguards, Training, Retention and Patient Data
    Handle a paused chat and model switching, understand the Life Sciences Verification Program (beta) and use cases, and apply the training, retention and patient-data rules for your plan.
  17. Level 17: Admin Outline and Troubleshooting
    Outline what organisation Owners and admins control, and diagnose the common errors a researcher meets while working.
  18. Level 18: Final Challenge: A Fully Traceable Analysis Exam
    Review the whole course by planning, running, tracing, reviewing and documenting one end-to-end analysis in a field of your choice while applying every control covered.

Access

The first 2 levels are free with a free account. Every level, the podcast edition and the AI tutor come with All Access at £4.99/month or any Creator plan — see pricing.

Start free →