Connectors

A connector is a standing link between a data source you own and a Divinci workspace. Once connected it ingests into a RAG vector (and optionally the fine-tuning pipeline), on demand or on a recurring schedule, without anyone re-uploading files.

Fourteen providers ship today. They differ in one way that matters more than any other — how you connect them — and that decides how much of the job the CLI can do.

The three connector shapes

Shape	Providers	Connect from
Config-only — the connector is its configuration; no secret to hand over	Web Crawl, WordPress	✅ CLI, API, web app
Credential — you paste a key/secret we validate before storing	AWS S3, Cloudflare R2, Google Cloud Storage, Recall.ai, Practice Better, Local Files	API, web app
OAuth redirect — the provider's own consent screen	Google Drive, Gmail, Google Calendar, Dropbox, Notion, Zoom	Web app (a human must click Allow)

Only two providers can be synced headlessly

divinci connectors sync and POST /api/v1/connectors/:connectionId/sync accept web_crawl and wordpress only. Every other provider's pipeline needs an explicit file selection and an interactive user token, so the API answers with a 400 naming the supported providers rather than queueing a job that could never pick files. Sync those from the Connectors page in the web app.

What the CLI does do for all fourteen: list, status, settings, sync-status, sync-cancel, disconnect and stats.

Every connector

Web Crawl Crawl a site, subpath or sitemap into a vector — recurring, with draft-and-swap.

WordPress Posts, pages and custom post types over the WP REST API.

Google Drive Documents and folders, with push notifications on change.

Gmail Email threads as assistant context.

Google Calendar Events and scheduling context.

Dropbox Files from a Dropbox account, including audio and video.

Notion Pages and databases from a Notion workspace.

Zoom Cloud recordings and transcripts.

AWS S3 Any bucket, scoped by key prefix.

Cloudflare R2 R2 over its S3-compatible API.

Google Cloud Storage GCS buckets via a service-account key.

Local Files Files pushed from a folder on your own machine.

Recall.ai Meeting recordings and transcripts via a bot notetaker.

Practice Better Clinical records, de-identified before indexing.

Managing connections list, status, settings, sync, sync-status, sync-cancel, disconnect, stats — the provider-agnostic commands.

Two things every connector shares

Pipelines (targetTypes)

A connection processes into rag, into fine-tuning, or both. Selecting fine-tuning always adds RAG — the SFT pipeline consumes RAG-processed chunks, so it cannot run alone.

The live-vector guard

If a connector's target vector is attached to a published release, chunk writes to it are rejected. The rejections are recorded, not thrown, so the sync reports success and nothing new becomes retrievable — the single most common "it ran but changed nothing" report.

Use draft-and-swap (--swap on the CLI, swapOnSuccess in the API/web form): the sync ingests into a fresh draft vector and atomically repoints your live release(s) at it only once the run succeeds. Both the web-crawl and WordPress forms default it on for exactly this reason.

Recurring sync

Set autoSync with a cadence in minutes (15-minute floor, default 1440 = daily):

divinci connectors settings wordpress --auto-sync on --cadence 1440

How a change reaches us differs by provider:

Mechanism	Providers	Latency
Push webhook	Google Drive, Gmail, Google Calendar (per-connection watch channels, auto-renewed), Zoom (app-level)	near real-time
Polling	everything else — and any webhook provider whose channel lapsed	the connection's cadence

A failed webhook registration degrades to polling rather than to silence.

See also

• RAG Knowledge Base — the vectors these connectors ingest into.
• MCP tool catalog — connector_list, connector_sync, connector_status and friends for agentic IDEs.
• divinci rag crawl — a one-shot crawl with no recurring connector behind it.