Organization settings¶
Organization settings are the shared operating environment for research: who belongs, which services and compute are available, what network access is allowed, and how usage is governed.
Open Settings and select the organization to see the current configuration. Members can inspect it; Admins and Owners can make changes. Treat those changes like infrastructure changes: understand who depends on the current behavior, make the smallest useful adjustment, and validate it with ordinary work.
Settings map¶
| Tab | Purpose |
|---|---|
| General | Organization identity. |
| Members | Roster, invitations, and roles. |
| Billing | Research Unit balance and usage. |
| Cluster usage | GPU and storage utilization for connected dedicated infrastructure. |
| Compute | On-demand compute and linked organization clusters. |
| Custom instructions | Persistent guidance applied across the organization. |
| Integrations | Shared Weights & Biases and Hugging Face connections. |
| Environment variables | Shared runtime keys and configuration. |
| Sandbox | Organization outbound-network policy. |
| Privacy | Provider retention mode and the organization's data-use preference. |
Members see restricted tabs as read-only. This visibility is useful when diagnosing why a workflow cannot reach a host or use a shared integration. Only the Owner can transfer ownership.
You do not need to configure every tab before the team begins. Start with membership, billing, and a ready compute option. Add shared integrations, variables, and policy as real workflows require them.
Custom instructions¶
Use organization instructions for validation, coding, safety, or reporting conventions that should be consistent across the team. Write them as standards a new researcher could understand without knowing the incident that inspired them.
Keep secrets and one-thread requests out of this shared layer. Review instructions periodically; an obsolete default can be harder to notice than a missing one because it continues to sound authoritative.
Integrations and variables¶
Use organization integrations only for shared service accounts. W&B and Hugging Face have dedicated cards; other runtime configuration belongs in Environment variables.
A personal environment variable overrides a same-named organization value. Keep that precedence in mind when only one member sees an authentication or behavior difference.
Configure GitHub under Personal settings → Integrations.
Sandbox¶
Sandbox controls outbound networking for Silico sessions and proxies work on unmanaged clusters. Use an allow-list for approved hosts.
Before tightening the policy, document the domains your workflows need and validate the change with a small experiment. This makes blocked dependencies much easier to diagnose. If a job suddenly cannot download a model or call an evaluation service, compare the requested host with this policy before changing code.
Privacy¶
Privacy shows whether model providers use zero data retention. This may limit available catalog models.
It also records the organization's preference for future data use. Goodfire does not currently train models on customer data.
