Combined

Combined

Permissioned company context for external AI agents.

Combined connects an Account's SaaS systems and employee-computer context, keeps the data fresh, and exposes it to external agents through read-only SQL, MCP, and typed SDKs.

The product owns acquisition, consent, provenance, freshness, grants, revocation, billing controls, and deletion. It does not build or host the agent consuming the data.

Choose your path

You are…Start with
Connecting an agentFive-minute connection
Setting up a company workspaceAccount onboarding
Operating Sources and accessProduct guide
Integrating over HTTPAPI overview
Using an SDKSDK reference
Configuring an MCP clientMCP guide
Installing Desktop CaptureDesktop setup
Diagnosing a failureTroubleshooting

Product surface

The authenticated application has four Account workspaces:

  • Sources connects SaaS systems and Desktop Capture, selects datasets and fields, schedules syncs, reviews schema changes, and controls lifecycle actions.
  • Query discovers the granted catalogue and runs bounded, read-only DuckDB SQL with receipts.
  • Access creates agents, grants exact Sources or datasets, applies limits, and revokes immediately.
  • Activity records content-free operational events, attempts, deliveries, authorization decisions, queries, deletion, backup, restore, and support access.

Account settings cover general identity, members, billing, usage, and privacy and retention. The Product guide explains each workspace and every operator-visible state.

Interfaces

Combined exposes one governed data path through the REST API, TypeScript SDK, Python SDK, MCP, and the in-product Query workspace. The API reference contains conventions and a generated page for every public REST operation. The MCP guide documents all six tools. Connector credential schemas and limitations are returned live by GET /v1/connector-definitions rather than copied into static prose.

One data path

The Query workspace, POST /v1/sql, MCP tools, and SDKs all use the same grant resolution, SQL validation, query isolation, and receipt path. No helper has a broader data surface than SQL.

Every queryable row can be traced to a Source, dataset, Run, Delivery, ingest time, and stable record identity. Query receipts contain operational metadata only; results are never retained in receipts.

Public and internal boundaries

These docs cover the supported customer and developer surface. Private service bindings, cron dispatch, webhooks, connector execution capabilities, infrastructure credentials, and recovery commands are intentionally not public APIs. Their security and operator runbooks remain inside the source repository rather than being presented as customer-callable endpoints.

On this page