Appearance
Sources
A Source is a connection from an AI System to an external system, with its credentials. Modulos uses it to read data and context from that system. Metric Tests, Agent Tests, Scout, and the Risk Agent read the Sources of the AI System. The Source type sets which of them can read a Source.
Where in Modulos
AI System → Settings → Sourceslists the Sources of the AI System and adds new Sources.Organization → Source & Connectors → Sourcesmakes each Source type available or unavailable for the organization.

- 1Sources tabThe Sources of an AI System are in its Settings.
- 2Add SourceAdds a Source to the AI System.
- 3ConnectionThe result of the last connection check.
Who can do what
Permissions
- Organization Admins make Source types available or unavailable for the organization.
- Owners of an AI System add, change, and delete its Sources in
AI System → Settings → Sources. Organization Admins, Risk Managers, and Policy Managers can do this in each AI System. - Editors and Viewers cannot open the Settings of an AI System and do not see its Sources there. Editors select Sources in Metric Tests, Agent Tests, and Scout.
Supported Source types
| Source type | Metric Tests | Scout, Agent Tests, Risk Agent | Connects to |
|---|---|---|---|
| AWS | ✓ | ✓ | CloudWatch metrics and resources of an AWS account |
| Azure | ✓ | ✓ | Metrics and resources of Azure subscriptions |
| Claude Console | ✓ | ✓ | A Claude API organization on platform.claude.com |
| Claude Enterprise | ✓ | ✓ | A Claude Enterprise organization on claude.ai |
| Datadog | ✓ | Metrics in a Datadog account | |
| GitHub | ✓ | A GitHub repository | |
| Langfuse | ✓ | ✓ | A Langfuse project |
| Microsoft Copilot Studio | ✓ | The Copilot Studio agents of a Power Platform tenant | |
| Modulos Client | ✓ | Metrics that your code sends to Modulos | |
| OpenAI Platform | ✓ | ✓ | An OpenAI API organization on platform.openai.com |
| Prometheus | ✓ | Metrics on a Prometheus server | |
| Snowflake | ✓ | ✓ | A Snowflake database |
| Vijil | ✓ | ✓ | Evaluation scores of an agent in Vijil |
Scout, Agent Tests, and the Risk Agent read the same Source types. The Risk Agent reads the active Sources of these types in the AI System. An Agent Test that has a Datadog or Prometheus Source in its Sources ends with an error.
For AWS, Azure, Claude Console, Claude Enterprise, Langfuse, OpenAI Platform, and Vijil, these AI agents read only one Source of each type in the AI System. Metric Tests can use each Source.
If a Source type is not available in AI System → Settings → Sources, the type is not available for the organization.
Make a Source type available
Each Source type is not available until an Organization Admin makes it available. See Source & Connectors, which also tells what happens to the Sources of a type that becomes unavailable.
Pull and push
- A pull Source: Modulos connects to the external system and reads the data. All types in the Add Source dialog are pull Sources.
- A push Source: your code sends data to Modulos. Modulos Client is the push Source. Each AI System shows it. It is active only when the organization activates Modulos Client.
Connection check
When you save a Source, Modulos tests the connection. If the test fails, Modulos does not save the Source. Most types then show a general error. The type pages tell you what Modulos checks.

- 1Source typeThe external system to connect to the AI System.
- 2FieldsThe credentials and settings of the selected type.
- 3SaveModulos tests the connection before it saves the Source.
After you save, the Connection column of AI System → Settings → Sources shows Healthy, Broken, Unknown, or Not applicable. Modulos checks again when you open the page and after you add, change, or delete a Source. Put the pointer on the status to see the cause. Modulos Client always shows Not applicable, because it receives data.
Keys and secrets
- Use a dedicated service account for each Source, with read access only.
- Replace keys and secrets on a schedule.
- After you save a Source, Modulos keeps its secrets and does not show them again.
Sources and Connectors
A Source belongs to an AI System, and all its members use it. A Connector belongs to one user and gives access with the rights of that user, usually through OAuth.
Sources vs connectors
Two trust models. Different scope, different lifecycle.
Sources
Sources
Service accounts attached to projects
- Shared across the project
- Used for automation and monitoring
- Designed for least privilege and rotation
Connectors
Connectors
User accounts connected to users
- Bound to a person
- Used for user-scoped access
- Reflects individual permissions and accountability
Why this matters
Different permission model
Different rotation and lifecycle
Different audit story
Related pages
Tests & Schedules
Metric Tests: one Source, one metric, one condition
Agent Tests
An AI agent examines the Sources against your Pass Criteria
AI System Settings
Where Sources are configured and who can change them
Scout Connectors
User-level OAuth Connectors for Scout
Integrations troubleshooting
Sources that fail validation, do not sync, or have no permission