How connectors work
The tools Pensieve reads, and how each one is read. Connect a source and its content becomes part of your company's map — cited, current, and scoped to exactly what you share.
A connector is a tool you let Pensieve read. Connect one, choose what's in scope, and its content flows into your context layer — filed onto the right pages, each claim citing the document it came from.
Connectors live under Data → Connectors in the app. Any member can connect a source; their credential joins the context's shared pool, so a single connection serves the whole team and colleagues needn't reconnect the same tool. Some connections require an admin or owner: Slack, Website and direct Postgres connections. Check the provider guide for its authentication and permissions.
Choosing what to share
Connecting a source isn't all-or-nothing. Each connector exposes a set of data types, and you pick the ones you want — HubSpot notes but not calls, Drive documents but not spreadsheets. Everything starts selected, so the default is simply all of it; untick what you don't want before you save.
You'll see the full list when you configure a connector, so the per-source pages below are the map rather than the menu.
Unticking later isn't the same as unticking before you've saved for the first time. Once something has been read, changing the selection can delete what it brought in. Read the removal rules below and any provider-specific exception before saving.
Things you add, and things you choose
Two of the connectors let you remove pieces individually, and the reason is what you did to add them.
With GitHub and Website you supply the thing itself — you type a web address, or you pick a repository and set its branch and how it should be read. Each one is a connection you made, so each can be disconnected on its own, with the same keep-or-delete choice as the whole connector.
Everywhere else you're choosing from what one authorisation already reaches: Notion pages, Drive files, Attio object types, Linear teams. Those aren't separate connections, so the way to remove them is to untick them — and unticking deletes.
How Pensieve reads a source
Not every source is worth reading the same way, and the difference is the spine of how Pensieve stays accurate. Literal sources are retained as documents, Distilled streams are compressed into durable signal, and Source Trees are understood as a connected structure. Those are connector readings, not billing categories: after conversion, retained spreadsheets use the lower-density Tabular representation towards your plan's footprint, whichever connector supplied them.
- Literal
- Each item is a deliberate artefact, so every one earns a source record and Pensieve's job is to compress it faithfully.
- Distilled
- An unbounded conversation with no natural unit. Pensieve drains it on its own schedule and keeps only durable facts — most of what it reads becomes nothing at all.
- Tree
- A single file means little alone; the structure is the knowledge. Pensieve researches the whole and maintains a Page tree for it.
| Connector | Reading | Scope | Stays current |
|---|---|---|---|
| Notion | Literal | Pages and databases — nested content included | Checked hourly |
| Google Drive | Literal | Files, folders, My Drive or shared drives | Checked every 30 minutes |
| GitHub | LiteralTree | Repositories — one branch and one reading each | Updates on every push |
| Supabase | Tree | Schemas — described as a database tree, never copied | Schema checked daily |
| Postgres | Tree | Schemas — described as a database tree, never copied | Schema checked daily |
| Slack | Distilled | Channels the bot has joined | On the schedule you set |
| Gmail | Distilled | Inboxes — each one connected and scheduled on its own | On the schedule you set for each inbox |
| Outlook | Distilled | Inboxes — each one connected and scheduled on its own | On the schedule you set for each inbox |
| HubSpot | Literal | CRM object types — their engagements follow | Checked hourly |
| Attio | Literal | CRM object types — their notes and transcripts follow | Checked hourly |
| Linear | Literal | Teams — their projects' overviews, updates and documents | Checked hourly |
| Pylon | Literal | Knowledge bases — requests and threads come workspace-wide | Checked hourly |
| Granola | Literal | Folders — subfolders included | Checked hourly |
| Fireflies | Literal | Every meeting the connected account can see | Checked hourly |
| Website | Literal | Public pages beneath the website addresses you add | Checked daily |
Pensieve only sees what you grant it
Across every source, the rule is the same: Pensieve reads what you explicitly share, and nothing more.
- Per-source scoping. Share specific Notion pages, specific Drive folders, specific Slack channels, chosen CRM object types — anything you don't grant stays invisible.
- Per-data-type scoping. Within a source you have connected, an unselected data type is never fetched at all.
- Per-context isolation. Each context gets its own map, in its own graph database. Nothing is shared across contexts.
- Read-only. Connecting a source lets Pensieve read it to build your map. It never writes back to your tools.
How a source reaches your map
Whichever way a source is read, the shape is the same:
- You authenticate, choose what's in scope — the pages, folders, channels, repositories or records — and choose which data types you want.
- Pensieve reads the material you selected, organises its evidence, and writes cited Pages. Connecting finishes before this work does.
- Follow progress on Data → Connectors. How it stays up to date explains when the first Pages become available and how to check processing.
- From then on it re-checks each source on a short cadence — half-hourly to hourly for most, daily for a website, and on every push for GitHub — re-reading literal sources, re-distilling conversational ones, and re-researching tree sources as they move.
All three readings land in the same place: one connected map of your company, ready to be read by your AI.
Check Data → Connectors for connection status and Data → Sources for individual documents. Each Page keeps its own History.
Sources: every document, listed
Open Data → Sources to inspect individual documents and their evidence. See Uploading and managing sources for supported files, upload limits, replacement, processing and deletion.
Up next
Notion