Customize your environment¶
Most work can begin with the defaults. Customize the environment when you notice something that should be reliably available beyond one thread: private context, a runtime key, a team convention, or additional compute.
The useful rule is simple: put enduring access and policy in settings; keep research-specific intent in the thread. This makes future work easier to reproduce without turning every conversation into configuration.
Personal and organization scope¶
Open the account menu and choose Settings:
- Personal settings apply to your work and credentials.
- Organization settings provide shared defaults and controls.
Use personal credentials for accounts you own and activity that should be attributable to you. Organization scope is a better fit for shared service accounts and team-wide defaults; changing it requires Admin or Owner access.
For example, connect your own GitHub account personally, but place a team W&B service account at organization scope when everyone should see the same runs.
Custom instructions¶
Use Custom instructions for evaluation, coding, safety, or reporting guidance that should carry across your work. Good instructions express durable standards: always report held-out sample size, prefer a particular formatter, or stop before using production data.
Keep one-off requests in their thread, where their context will remain clear. “Use this dataset for the next run” is research intent, not a standing instruction.
Personal instructions apply to your threads; organization instructions set the shared default. Never include secrets.
Integrations¶
Personal integrations include:
- GitHub for browsing and attaching repositories you can access
- Weights & Biases for training runs and metrics
- Hugging Face for private or gated models and datasets
Organizations can share W&B and Hugging Face connections. GitHub remains personal.
Public resources need no connection. For private material, use the smallest token scope that lets Silico do the work. If a connection fails later, the docs and thread record should still make clear which external resource the experiment expected.
Environment variables¶
Use Environment variables for runtime configuration such as API keys. Values are masked after saving.
A personal value overrides a same-named organization value. This is useful for testing your own credential, but it can also explain why a workflow behaves differently for you than for a teammate. Rotate shared values carefully and record operational ownership outside the secret itself.
Never paste a secret into a thread, plan, report, repository, or support message.
Storage¶
Personal settings → Storage groups experiment files by organization, folder, thread, and experiment. Choose Refresh, then expand the hierarchy to find what uses space.
Storage cleanup is permanent and can make an older report harder to inspect or reuse. A report may remain readable after its intermediate files are removed, but a future audit or follow-up may not. Confirm the exact scope and any dependent work before deleting files.
Compute¶
On-demand compute is ready for most organizations. Admins and Owners can link more options under Organization settings → Compute. Cluster usage shows dedicated infrastructure utilization.
See Compute for the complete workflow.
Security and privacy controls¶
Admins and Owners manage two organization policies:
- Sandbox controls outbound networking.
- Privacy shows provider retention and the organization's data-use preference.
Members can inspect the current policy but cannot change it. See Organization settings.
Next step¶
Continue with Working with a thread.