Connect Google Drive to an AI Knowledge Base

Uploading files is the slowest way to feed a knowledge base, because the moment you upload, the copy starts aging. Connect Google Drive, GitHub, or ClickUp instead — or crawl a public site — and the source stays current. Here is how, and how to read the sync status.

Most people meet a knowledge base by uploading a file. It works, and for a one-off document it is exactly right. But upload is a snapshot: the instant a PDF lands in the base it stops tracking the original, so six months later the agent is confidently citing a policy someone revised in March.

The fix is to stop copying and start connecting. An Insulin knowledge base can crawl a public website, or connect GitHub, ClickUp, or Google Drive — pulling from the systems your team already keeps up to date, and re-syncing as they change. This post is the how-to: what each connector actually brings in, and how to read the status so you know a source is really in the base.

How do I connect Google Drive to an AI knowledge base?

You connect Google Drive to an Insulin knowledge base by adding it as a source and selecting the folders to sync, rather than uploading each file. Insulin then pulls the documents in those folders into the base and keeps them in step as they change.

Drive is the connector most teams reach for first, because most of the material worth grounding an agent in already lives there — playbooks, policy docs, templates, runbooks. Point the connector at the folder that holds them and the base fills itself. The Google Drive connector is available for organization-level knowledge bases, which is the level you want anyway when the whole team should read the same material.

The distinction that matters: a folder you connect is not a folder you copied. When someone edits the pricing policy in Drive, the connected source re-syncs, and every agent attached to that base answers from the new version. Nobody re-uploads anything.

What does the GitHub connector actually sync?

The GitHub connector syncs a repository’s issues and pull requests as Markdown — not its source code. You pick the repositories and toggle which content types to include, and Insulin brings that discussion into the base.

That scope is deliberate, and it is the useful part. The decisions your team argued out sit in issues and PR threads: why a design was chosen, what a fix actually addressed, which edge case bit you last quarter. Source code answers what the system does; the issues and PRs answer why, and that context is exactly what an ungrounded model cannot reconstruct. Connect the repositories where those conversations happen and an agent can answer questions a code search never could.

Because it is a connector and not an export, a closed issue or a merged PR keeps flowing in as the work continues — the base stays a live record rather than a dump you took once.

Can I connect ClickUp, or crawl a website?

Yes to both. ClickUp connects as a source too, syncing tasks and documentation pages from the spaces you select — useful when your operating knowledge lives in ClickUp docs rather than in files. Website crawling takes a public URL, a crawl depth from 1 to 10, and imports up to 100 pages of content.

Crawling is the right tool for material that is already published and already the canonical answer — your own help centre or public documentation. You give it the starting URL, set how many links deep to follow, and it ingests the pages so an agent can cite them the same way it cites an uploaded PDF.

Two quick choices to get right:

Depth is not “more is better.” A depth of 1 or 2 from a docs index usually captures the real content; crank it to 10 and you also pull in navigation, tag pages, and whatever the site links out to.
Crawl what you don’t control the source of; connect what you do. If the pages sit in a Drive folder or a repo you can connect, connect it — you get re-sync. Crawl is for the published surface where a connector isn’t an option.

How do I know a source actually made it in?

Every connector and every document shows its sync state, so you can see a source is ready rather than assume it. A document moves through pending → indexing → indexed, or lands on failed if something went wrong — and connectors and crawls show a progress bar with a percentage alongside a running count of documents indexed.

This visibility is the whole reason connecting beats a blind bulk upload. A file that failed to extract does not announce itself in an answer; it simply is not there, and the agent gives a thinner reply with no explanation. Watching the status turns that silent gap into something you can see and fix.

Two habits worth forming:

Confirm the count after a sync. When a connector finishes, check that the number of indexed documents matches roughly what you expected from that folder, repo, or crawl. A large gap means something failed quietly.
Read a failed state as information, not noise. A failed document is usually a scan with no real text, or a file over the 25 MB ceiling. Fix the source, not the base.

This is what shipped as organization-level connectors — see the Insulin changelog for the GitHub and Google Drive connector launch, which added per-connector and per-document sync status for exactly this reason.

Connect it, then attach it

Connecting a source and using it are two steps. The connector fills a knowledge base; the base only shapes an answer once it is attached to an agent. An agent reaches only the bases you attach to it, so connecting Drive to the deal-desk base does not leak it to the finance agent — scope stays where you put it.

The payoff of connecting over uploading shows up here. Because the source re-syncs, the grounding an agent works from is the current version — in an interactive chat, in a shared channel, and in an unattended overnight job alike. Update the policy in Drive, and every one of those surfaces answers from the new version without anyone touching the base.

Frequently asked questions

How do I connect Google Drive to an Insulin knowledge base? Add Google Drive as a source on an organization-level knowledge base and select the folders to sync. Insulin pulls those documents into the base and re-syncs them as they change, so you never re-upload.

What does the GitHub connector bring in? Issues and pull requests as Markdown, not source code. You choose the repositories and toggle which content types to include, which brings in the discussion and decisions behind the work.

Can a knowledge base crawl a public website? Yes. Give it a public URL and a crawl depth from 1 to 10, and it imports up to 100 pages. Crawling suits published material like your own help centre or documentation.

Should I upload a file or connect the source? Connect anything that changes. An uploaded file is a snapshot that keeps being cited after the original is revised; a connected source re-syncs, so agents answer from the current version.

How do I know a connected source finished syncing? Each connector and document shows its state — pending, indexing, indexed, or failed — and connectors show a progress bar with a running count. Confirm the indexed count matches what you expected.

Takeaways

Uploading a file is a snapshot that ages; connecting a source keeps it current. Connect what changes.
Google Drive connects by folder selection on an organization-level base — the fastest way to load the material you already maintain.
The GitHub connector syncs issues and pull requests as Markdown, not code, which is where the why behind the work lives.
ClickUp syncs tasks and docs from selected spaces; website crawling takes a public URL, a depth of 1–10, and up to 100 pages.
Every connector and document shows sync state — pending, indexing, indexed, or failed — so a silent gap becomes something you can see and fix.

Insulin knowledge bases connect the systems your team already uses and let agents cite what they used. Explore Insulin knowledge bases, see how agents are scoped, or get a demo.