Environment Variables

Environment variables are pushed to every sandbox in your org and exposed to your run commands. Use them for API keys, model credentials, or anything else your scripts need at runtime — without committing secrets to your repo.

Scopes

Env vars can be set at two scopes:

• Org-level — set in Org settings > Environment variables. Available to every project in the org. Use this for shared credentials (e.g. WANDB_API_KEY, HUGGINGFACE_TOKEN) that the whole team uses.
• Project-level — set in Project settings. Only available to that project's sandboxes. Use this for project-specific config.

When the same key is set at both scopes, the project value wins. Override at the project level when one project needs a different value than the org default.

How they're delivered

OpenResearch writes the merged set to ~/.openresearch/env on each sandbox and sources it before every run. Your scripts see them as ordinary environment variables, so SDKs that auto-read them (wandb, HuggingFace, OpenAI, etc.) just work — no flag-threading or code changes needed.

Changes sync to every affected sandbox automatically — add or edit a key and your next run picks it up. No restart needed.

Reserved keys

OpenResearch sets IS_SANDBOX=1 on every sandbox. Don't override it — your scripts can use it to detect whether they're running inside OpenResearch versus locally.