Skip to main content
Connectors bring outside data into the vault and turn it into typed memory. Three layers ship today.

​
Layer 1: substrate ingest skills

Slash-command skills that pull data on demand. Output goes to External Inputs/<Source>/<scope>/<YYYY-MM-DD>.md. Idempotent: re-running on the same day overwrites cleanly. These six are the open-source default set. Each one ships in the ai-brain-starter substrate.

​
Layer 2: webhook receivers (paid runtime)

Real-time push from the source. The runtime exposes per-tenant signed URLs at POST /webhooks/<source>/{tenant_id}. Each source has its own signature scheme and per-tenant signing secret. The receivers covered in v1 of the runtime:
  • Slack, GitHub, Notion, Linear, Gmail, Salesforce
  • Microsoft 365, Confluence, Google Workspace
  • ServiceNow, Workday, SAP
  • Box, Snowflake, Databricks, Oracle
  • Jira, HubSpot, GitLab, Dropbox, Zoom, ADP
  • WhatsApp (in-process bridge, not HTTP)
Twenty-three sources total. See API Reference / Runtime for the full URL + signature scheme table.

​
Layer 3: enterprise connectors (Round 3, 2026)

Seven connectors added for F100 procurement requirements. Each one supports SAML SSO + SCIM provisioning + audit-log forwarding to the customer’s SIEM. The Round 3 set covers the connectors that platform-team buyers ask for first when evaluating an AI runtime. List and integration notes ship here as each one stabilizes.

​
Adding a connector

Two paths:
  1. Substrate skill (open source): add a new ingest-<source> skill to ai-brain-starter following the pattern in skills/_shared/. PR welcome.
  2. Webhook receiver (paid runtime): receivers are added in the runtime repo. Open an issue describing the source, signature scheme, and event shape, and we will scope.