Skip to content

Science Workspaces ​

Science Workspaces provide a persistent notebook-style environment for organizing GhostScience computations. Each workspace contains ordered cells, supports branching for experimentation, and can optionally link to a Federated Clean Room for multi-org computation.

Creating a Workspace ​

Create a workspace via the API or SDK:

python
from ghostscience import GhostScienceClient

client = GhostScienceClient(api_key="gst_...")
ws = client.create_workspace(name="Drug interaction PK/PD model")

Each workspace starts with a main branch and no cells.

Cells ​

Workspaces contain ordered cells of different types:

Cell TypeDescription
computationPython code executed in the PhD sandbox
markdownDocumentation and notes (not executed)

Adding Cells ​

python
client.upsert_cell(
    workspace_id=ws["id"],
    cell_id="cell-1",
    cell_type="computation",
    content="import sympy as sp\nx = sp.Symbol('x')\nsp.integrate(x**2, x)",
)

If cell_id is omitted, one is generated automatically. Upserting an existing cell ID updates the cell in place.

Running Cells ​

python
result = client.run_cell(workspace_id=ws["id"], cell_id="cell-1")
print(result["output"]["stdout"])

Computation cells execute in the PhD sandbox. Output includes stdout, stderr, exit_code, execution_time_ms, and backend. Markdown cells are skipped.

Cell Dependencies ​

Cells track dependencies via depends_on (a list of cell IDs). When an upstream cell changes, downstream cells are marked is_stale: true so you know they need re-execution.

Branching ​

Create branches to explore alternative approaches without affecting the main workspace:

python
client.create_branch(
    workspace_id=ws["id"],
    name="experiment-adam-optimizer",
    parent_branch="main",
)
branches = client.list_branches(workspace_id=ws["id"])

Branches share the same cells table but provide a named reference point for experimentation.

Clean Room Mode ​

A workspace can be linked to a Federated Clean Room for multi-org computation:

python
ws = client.create_workspace(
    name="M&A Revenue Overlap",
    clean_room_mode=True,
    clean_room_id="cr-abc123",
    participating_org_ids=["org-1", "org-2"],
)

When clean_room_mode is enabled and more than one organization participates, the workspace operates under clean room governance: attestation, escrow, and dual-approval release apply to all computation cells.

Access Control ​

  • Owner -- the user who created the workspace has full access
  • Organization -- other users in the same org can view and edit
  • Cross-org -- clean room workspaces are visible to all participating orgs

SOC 2 Type II Certified | BAA Available | HIPAA-Ready