Skip to content

Discovery Ingestion Sources

Discovery ingestion lets Solutioneer pull source material from the tools your team already uses. The primary consumer is discovery-context-ingester, which normalizes everything into one reusable fact sheet.

SourceConnectionTypical use
HubSpotComposio managed authCRM records, contacts, and deal context
SlackComposio managed authChannel history, threads, and message search
GmailComposio managed authCustomer email threads and attachments
Google DriveComposio managed authShared docs, sheets, and file parsing
GranolaPreconfigured MCP auth onlyMeeting transcripts, when auth is already wired up

All five sources are read-only and opt-in. Codex does not ingest from any of them unless the user explicitly asks.

Four of the five sources share the same connection model:

  • Managed auth via Composio. hubspot, slack, gmail, and googledrive use Composio’s hosted OAuth flow. The first time a skill reaches for one of them, Codex presents a Connect now / Skip choice.
  • Connect now opens the Composio hosted login page in a browser tab when possible. Codex waits for the connection to become ACTIVE, then proceeds.
  • Skip keeps the workflow chat-only. Unresolved private claims are labelled unknown.

Granola is different:

  • granola_mcp requires MCP auth that is already configured in your environment. Solutioneer does not attempt to auto-configure Granola.
  • If Granola auth is not ready, Codex offers a manual transcript fallback: paste the transcript or point at a local file.

CRM auto-ingest is fixed to HubSpot. For Salesforce or any other CRM, export the relevant records as CSV, paste records into chat, or save them to a local file — discovery-context-ingester will normalize those the same way.

Ingested claims carry a composio://<toolkit>/... or local://... provenance marker so downstream skills can trace evidence back to the source. Any claim that cannot be grounded is labelled as an assumption or unknown rather than invented. See Grounding Rules for the full policy.