Retrieval.

Ingestors cover knowledge sources, ticketing, CRM, code, files, observability, and data warehouses. New ingestors land weekly. Data ingestors are distinct from MCP servers, which power Apps and Agent tool calls. Many vendors offer both. For document stores (SharePoint, OneDrive, Google Drive, Box, Documentum, S3) the ingestors index file titles and metadata today; search inside file contents for those sources is in development.

A source ACL is translated into the engine’s own allow-list. Some sources express permissions in shapes an allow-list cannot represent: a conjunction, an inherited hierarchy, or a rule that can only be resolved at query time. For those documents Retrieval withholds the document rather than indexing it with a wider ACL, because the alternative is a leak. The connector logs a warning at startup and reports the withheld count on every run, so the shortfall is visible rather than silent. Query-time permission checks are available for sources that need them; ask which of your sources are covered. A source without them may index only part of its corpus.

Beyond the in-house ingestor catalog, the Airbyte bridge extends Retrieval to hundreds more SaaS and database sources. Anything in the Airbyte ecosystem can become a Retrieval source.

Keyword, vector, and entity-graph traversal in one query. The engine picks the right mix per query. Customers do not have to choose a search mode; they get the best of all three.

Relevance, recency, source quality, trust signals: all shapable through policy code. The engine ships sensible defaults; customers extend with their own rules. Authorization stays platform-enforced and is never customer-overridable.