GovernSecurity
Credential Management
Centralized management of API Key / OAuth Token / Cookie · Encryption / rotation / audit
A Credential = sensitive information used by Evose calling external services. Centralized management + field-level encryption + audit + fine-grained authorization.
Credential Types
| Type | Use |
|---|---|
| API Key | SaaS API calls (OpenAI / SendGrid / internal APIs) |
| OAuth Token | User-authorized third-party access (Google / Feishu / DingTalk) |
| Cookie / Session | Sites that require a logged-in session |
| Custom | SAML / custom headers / multi-field combinations |
Security Mechanisms
| Mechanism | Description |
|---|---|
| Field-level encryption | AES-256 in DB |
| Decrypt on use | Decrypted only at call time, never persisted |
| Audit logs | Who / when / which tool / which Agent used it |
| Rotation support | One-click replace; smooth swap of old/new |
| Forced update notifications | Notified proactively on near-expiry / leak |
Register a Credential
Reference in Tools
In tool config, use placeholders, never plaintext:
Usage Tracking
The credential detail page shows:
- Which tools / Agents / Workflows reference it
- The most recent N call records
- Call source (workspace / user)
→ Look here before rotating to assess impact.
Rotation Flow
- In the external service (e.g. OpenAI console), generate a new key
- In the Evose credential detail → Rotate → input the new value
- The system swaps gradually: new calls use the new key; in-flight old keys finish naturally
- After 24 hours, confirm no errors → fully retire the old key
Credential vs API Key Management (Layer 1)
Easy to confuse:
| Credential (this page · Layer 3) | API Key Management (Org · Layer 1) | |
|---|---|---|
| Direction | Evose calls external | External calls Evose |
| Examples | OpenAI Key, SendGrid Key | The Evose API Key you give the customer system |
| Encryption | Field-level encrypted | Same secure handling |
| Audit | Whose credential was used | Who called the Evose API |
Anti-Patterns
- Multiple Agents share one key without per-workspace separation — hard to track, hard to rate-limit
- Credentials written into "test URL" of the tool — URL ends up in logs
- Production key reused in dev workspace — distinguish with
_prod_dev
Next Steps
- Configure resource policies → Resource policy ACL
- Tools referencing credentials → Workspace · Tools