Grounding Agentforce on SharePoint and Confluence: Unified Knowledge is retiring, and what replaces it

Here’s a problem that doesn’t show up in any Agentforce demo: the knowledge your agent needs to answer from doesn’t live in Salesforce. Your support runbooks are in Confluence. Your product documentation is a SharePoint site. Your policy PDFs sit in a Google Drive folder three teams share. The Data Library tutorials assume your content is already in Salesforce Knowledge or a pile of PDFs you can upload — but yours isn’t, and nobody’s going to migrate a thousand living Confluence pages into Knowledge articles by hand.

The bridge Salesforce built for exactly this was Unified Knowledge — a set of connectors that reach into SharePoint, Confluence, Google Drive, and your websites, and sync those articles into Salesforce so an agent can ground on them. It works. And it’s on the way out: Unified Knowledge is scheduled to retire with the Summer ‘27 release (the retirement window has been reported as June 2027 — verify the exact date against Salesforce’s live retirement notice before you plan around it, because these move). The replacement isn’t a rename. It’s Salesforce Enterprise Knowledge, powered by Data 360, and it grounds an agent through a genuinely different mechanism.

If you’re standing up external-knowledge grounding today, this is the decision you can’t afford to get wrong: build on the thing that’s retiring, or the thing that replaces it. This post is what each one actually does, how the grounding chain works end to end, and a move that doesn’t strand your agent mid-transition.

One naming note first, because it’s in every doc: Salesforce rebranded Data Cloud to Data 360 in October 2025. The successor’s launch materials say “powered by Data Cloud”; current docs say “powered by Data 360.” Same platform, newer name — read them as synonyms, and see the full rebrand explainer if you want the history.

What Unified Knowledge actually does

Unified Knowledge was built through a partnership with Zoomin, and the Zoomin lineage explains its architecture. You configure a connector in Setup — point-and-click — to a third-party source: SharePoint, Confluence Cloud, Google Drive, websites, and others like Jira or custom content. Salesforce then imports and syncs those articles into Salesforce in a read-only format — HTML or PDF — and reindexes them from the source multiple times a day so the content stays reasonably current.

The word that decides everything downstream is read-only. Unified Knowledge doesn’t move authorship into Salesforce; it makes a synced, read-only copy of what lives in the source system, so support reps and customers get one integrated search experience, and Einstein for Service’s generative features have something governed to ground on. You still edit the Confluence page in Confluence. Salesforce holds a mirror.

That model has real strengths — your authors keep their tools, the sync is hands-off — and one honest limitation: it’s a mirror of articles, imported as HTML/PDF, not a harmonized knowledge layer. Different sources come in looking like their source. And the connector set, sync behavior, and article limits are exactly the kind of detail that shifts between releases, so confirm the current Unified Knowledge Limitations and Considerations pages before you design around any specific cap.

On cost: Unified Knowledge shipped with a 90-day free trial that includes three connector instances, tied to Service Cloud Unlimited, Einstein 1 Service, or the Knowledge add-on. That trial is a Unified Knowledge fact — don’t assume it carries over to the successor, which rides on Data 360’s own entitlements.

Why the successor is a re-platform, not a rename

Enterprise Knowledge, powered by Data 360, solves the same business problem — get external knowledge in front of an agent — but it does it through Data 360 instead of a Zoomin importer, and that changes the mechanics.

Instead of syncing read-only article copies, Enterprise Knowledge unifies native Salesforce Knowledge with external knowledge ingested through Data 360 and harmonizes it — rendering articles in a single consistent format regardless of origin, viewed through a new content-viewer experience. More importantly for an agent builder, it establishes grounding for Agentforce by creating a Data 360 search index and a retriever over that content. External sources come in through Data 360’s unstructured ingestion — a web crawler, a connector — landing in a Data Lake Object, which is then indexed and retrieved.

That’s the same Data 360 RAG machinery that grounds any agent on unstructured content — the search index, the chunking, the retriever — rather than a knowledge-article mirror. If you’ve read our Data Library walkthrough, this will click: Unified Knowledge got external content into Salesforce Knowledge; Enterprise Knowledge gets it into the Data 360 retrieval layer your agent already grounds on. The destination is different, and so is what you can tune once it’s there.

Enterprise Knowledge reached general availability in 2026 (reported around the Summer ‘26 wave — confirm the exact GA state for your edition against the live docs, as this is version-dependent). The point for planning: the successor is available now, well ahead of the retirement, which is the whole reason you have runway to move deliberately instead of scrambling in 2027.

How external knowledge actually grounds an agent

Whichever product feeds it, the last mile to the agent is the same retrieval chain, and getting it exact saves you a day of “why is my agent ignoring the docs” debugging.

1. Content is ingested and chunked. External articles or documents land in Data 360 (a Data Lake Object for unstructured content), where they’re split into passages.
2. Chunks are embedded into a search index. Data 360 builds a search index over the vectors. This index supports vector search and hybrid search — vector plus keyword — which matters when your content has exact part numbers or error codes that pure semantic search fluently misses.
3. A retriever queries the index. A retriever is the resource that searches the knowledge store and returns the relevant passages. You can scope it to a subset of the content so the agent doesn’t retrieve from the wrong corpus.
4. The agent calls it. In Agentforce, the retriever backs the Answer Questions with Knowledge action (or a custom prompt template). At runtime the action runs the search, pulls the top passages, and grounds the model’s answer on them — retrieval-augmented generation, end to end.

The single most common failure here isn’t the model — it’s step 2 not finishing. The search index has to reach a Ready status before the action returns anything. Attach a knowledge source to an agent while the index is still building and the agent retrieves nothing, which reads exactly like a broken action but is really an unfinished index. Wait for Ready, then test.

The second most common failure is permissions. The retriever runs as the agent’s running user, and that user needs access to the underlying data — the same running-user identity trap that catches custom actions. It works in your test because you can see everything; it returns nothing at agent runtime because the bot user can’t.

Where this sits next to Data Library and custom RAG

Three grounding surfaces get conflated constantly. Keeping them straight is what tells you which one to reach for:

• Agentforce Data Library — the fast, packaged path. It auto-provisions the whole RAG stack (stream, objects, vector store, search index, retriever, prompt template, action) with the knobs hidden and set to defaults, over Salesforce Knowledge, uploaded files, or web search. It’s the on-ramp, and often all you need. It does not, on its own, reach into SharePoint or Confluence.
• Enterprise Knowledge, powered by Data 360 — the successor to Unified Knowledge specifically for external knowledge systems, harmonized and grounded through Data 360’s index and retriever. This is the answer when your source of truth lives outside Salesforce and you want it unified with native Knowledge.
• Hand-built Data 360 RAG — the granular path. You build the search index and retriever yourself, choosing chunking, hybrid search, top-k, and filters, when the defaults retrieve the wrong passage and you can point at the questions they get wrong.

They’re not competitors so much as a ladder. Start packaged; graduate to explicit Data 360 control when a real retrieval failure justifies it. Enterprise Knowledge is the piece that fills the specific gap — external systems — that a plain Data Library and a folder of uploaded PDFs don’t cover.

A migration sequence that doesn’t strand your agent

If you’re on Unified Knowledge today, or about to start, the order of operations matters more than the effort. Working back from the retirement:

1. Confirm your Data 360 foundation. Enterprise Knowledge grounds through Data 360 — so Data 360 has to be provisioned, and external content has to be ingestible into it (crawler or connector). If Data 360 is already carrying other use cases, you’re mostly wiring knowledge onto an existing foundation. If it isn’t, standing up that foundation is the real project with the lead time — the grounding wiring on top is fast.
2. Inventory every source and how the agent uses it. List each SharePoint site, Confluence space, and Drive folder you sync today, which agent topics and actions depend on it, and how fresh it has to be. This is your parity checklist.
3. Re-ground on Data 360, one source at a time. Ingest a source into a Data Lake Object, build the search index, wait for Ready, attach a retriever, and point the Answer Questions with Knowledge action at it. Verify the running user can see the data. Test with questions a real customer asks — not the one whose answer sits in a single clean paragraph.
4. Run both in parallel during the cutover. You don’t have to flip everything at once. Ground the agent on the new Data 360 retriever for migrated sources while Unified Knowledge still serves the rest, and move sources across as you validate retrieval quality. The retirement is your deadline, not your cutover date — aim to finish with room to spare.
5. Re-measure retrieval quality, don’t assume parity. A re-platform is exactly when someone notices the agent answers a question differently than before. Because the new path is a real search index you can tune, that’s an opportunity — hybrid search, better chunking, scoped retrievers — not just a risk. But test it, because a confidently wrong grounded answer looks identical to a right one from the chat window.

The teams that will be calm in 2027 are the ones who confirmed the Data 360 foundation and ran the source inventory early, because those are the two steps with real lead time. Re-grounding a source, once the foundation exists, is an afternoon.

Takeaways

• The problem is real and specific: your agent’s knowledge lives in SharePoint, Confluence, and Drive, not in Salesforce, and the standard Data Library path doesn’t reach it.
• Unified Knowledge is retiring with the Summer ‘27 release (verify the exact date on Salesforce’s live notice). It’s a Zoomin-built importer that syncs external articles into Salesforce as read-only HTML/PDF, reindexed multiple times a day.
• The successor is Enterprise Knowledge, powered by Data 360 — not a rename. It grounds an agent by ingesting external content into Data 360 and building a search index + retriever, harmonized with native Knowledge, rather than mirroring read-only copies.
• The grounding chain is the same last mile: ingest → index (vector or hybrid) → retriever → the Answer Questions with Knowledge action. The index must reach Ready, and the agent’s running user must have access, or you’ll debug an empty result as a model problem.
• Migrate deliberately: confirm the Data 360 foundation, inventory every source, re-ground one at a time, run both in parallel through the cutover, and re-measure retrieval quality — the runway to Summer ‘27 is the point.

Getting an agent to answer accurately from knowledge scattered across a dozen systems — and knowing whether a wrong answer is your grounding, your chunking, or the running user’s permissions — is exactly the Agentforce grounding work we do. If Unified Knowledge’s retirement is on your roadmap, the time to plan the move is now, while you still have runway.