Credentials for Agent tools
Credentials are the API keys, tokens, connection strings, and other private values an Agent needs to use external services.
How credentials work
SandBase stores each credential as an encrypted, write-only value. The Console and public API expose its name, scope, status, and a masked hint—not the saved value. At execution time, the runtime resolves active credentials that match the Agent or Service configuration.
Managing credentials
In the Console
Navigate to Developer → Credentials.
When adding a credential, choose the narrowest useful scope:
| Scope | Use it for |
|---|---|
| Workspace | A value available to eligible Agents in the current workspace |
| Agent | A value associated with one Agent ID |
| Service | A value available to a published Service workflow |
Saved values cannot be read back. To replace one, use the credential's Manage action and rotate it. A credential can be disabled; disabled records remain visible as metadata.
Common Variables
| Variable | Example | Purpose |
|---|---|---|
GITHUB_TOKEN | ghp_xxxx | Authenticate with GitHub API |
DATABASE_URL | postgres://... | Connect to a database |
SLACK_WEBHOOK | https://hooks.slack.com/... | Post notifications |
OPENAI_API_KEY | sk-... | Call OpenAI directly from a tool |
Security
- Values are encrypted before storage and omitted from credential responses
- The Console shows only masked metadata after creation
- Scopes control where a credential can be selected; use Agent scope when a workspace-wide value is unnecessary
- Keep secrets out of prompts, instructions, test input, and manually authored event payloads
Next steps
- Define an Agent — create an Agent that uses credentials
- Tools and credentials — capabilities that consume these secrets
- Services — publish an Agent with its credentials available at runtime