Skip to main content
Version: 0.9.19

Data connectors

A data connector links an agent to one of your data sources so it can work with your data at runtime. This page covers the connector types, how to create and test each one, the quality gates that make a connector usable, versions, and triggers.

Connector types​

AlphaAgent supports five connector types:

TypeConnects to
SQLA PostgreSQL or MySQL database
REST APIAn HTTP API described by an OpenAPI 3.x specification
MCPAn external tool server that speaks the Model Context Protocol
SnowflakeA Snowflake data warehouse
AWSAn AWS data source, accessed by assuming a role you create

Once created and wired to an agent in the Agent Map, an agent can reach the connector's data while it runs code. An agent can use up to 8 connectors. For how credentials are handled at runtime, see Connector data in the sandbox.

Creating a connector​

From the Data Connectors list, click Create. Every connector has:

  • Name and description.
  • Connector type (above).
  • A connector guide — text describing what agents can do with this connector. The guide helps an agent use the connector well; for most types AlphaAgent helps you generate it (see below).

Nothing is saved until you click Create Connector at the end — you can test and refine freely first.

[Screenshot: Create connector form]

Quality gates by type​

Each type has steps you must complete before Create Connector is enabled. These gates make sure a connector actually works before agents rely on it.

SQL and Snowflake​

  1. Enter the connection details — for SQL: database type, host, port, database name, and an optional schema; for Snowflake: account identifier, warehouse, and database name.
  2. Enter credentials. Snowflake offers an auth method selector: key pair (with optional passphrase), token, or password.
  3. Click Test Connection.
  4. Click Explore & Enrich. AlphaAgent inspects the schema and streams back an enriched connector guide.
  5. Review the enriched guide and click Accept & Use This Guide.

REST API​

  1. Enter the base URL and paste the OpenAPI 3.x specification (raw JSON).
  2. Add credentials as key/value rows (each key maps to a header or query-parameter name), and choose a credential placement — custom headers or query parameters.
  3. Click Validate OpenAPI. Validation must pass.
  4. Generate an enriched guide from the spec, review it, and Accept it.

MCP​

  1. Enter the MCP server URL (streamable HTTP). Credentials are optional, entered as key/value rows.
  2. Click Test to run a draft connection.
  3. Run tool discovery + enrichment, review the enriched guide, and Accept it.

AWS​

AWS connectors use role assumption — there are no stored passwords, and the guide is written by you (there's no enrichment step):

  1. Click Generate Role Setup. AlphaAgent returns the principal to trust, a unique External ID, and a ready-to-paste trust policy — each in a copyable block.
  2. In your AWS account, create an IAM role using that trust policy.
  3. Paste the role ARN back into Studio and run the AssumeRole test. A successful test is the gate to create.

Versions​

Each save creates a new immutable version of the connector. Agents and workflows pin to a specific version, so older versions stay available until everything that uses them has migrated. The connector's detail screen shows a versions table with the current version highlighted.

To change a connector, use Update — it mirrors the create form, pre-filled with the current values, and publishing saves a new version.

Viewing a connector and its dependents​

A connector's detail screen shows its configuration (with stored secrets masked, and per-field reveal toggles) and the DC Map — the agents whose active configuration uses this connector. As with other resources, you can Disassociate All and Re-associate All to manage dependents in bulk, and deleting is blocked by an impact guard if anything depends on the connector.

Triggers​

Triggers let a connector watch a data source on a schedule and start a workflow when a condition is met. They're shown in a panel on the connector's detail screen.

[Screenshot: Connector triggers panel]

To create a trigger, set:

  • A name.
  • A poll interval in seconds (minimum 30; default 300) — how often the connector is checked.
  • A condition type:
    • Deterministic — a precise JSONPath expression with a comparator and threshold.
    • Natural language — a free-text description for fuzzy intent matching.
  • The workflow to run when the condition matches.

When the condition is satisfied, the chosen workflow starts automatically.