On this page

Put Coworker to work on your stack.

Connect Salesforce, Slack, Jira and run your first agent in minutes.

Book a demo
Blog

Enterprise AI

Glean Knowledge Graph: How It Works and How It Compares to Coworker OM2

How the Glean Knowledge Graph and Enterprise Graph work, what they power, their limits, and how Coworker AI's OM2 company graph compares.

Dhruv Kapadia10 min read

The Glean Knowledge Graph is the model of your company that Glean builds from everything its connectors index: content, people and activity, mapped together and kept up to date. Glean uses it to rank search results, ground Glean Assistant's answers and give Glean agents context, and since September 2025 it sits under a broader Enterprise Graph that adds business entities and personal graphs.

I work at Coworker, which builds a competing company knowledge graph called OM2, so read this with that in mind. Everything I say about Glean comes from Glean's own docs and product pages, linked below and read on November 9, 2026. Glean's graph is a serious piece of engineering, and for some companies it is the better choice. I say which at the end.

For background on the concept, see my explainers on enterprise knowledge graphs and context graphs, and my roundup of knowledge graph software.

What is the Glean Knowledge Graph?

Glean describes its Knowledge Graph as the layer that "organizes your enterprise's indexed information in real time" (Glean docs). It rests on three pillars:

PillarWhat it covers (Glean's description)Why it matters
ContentDocuments, messages, tickets and other content across the companyWhat the company knows
PeopleUser identities, roles, teams and organizational relationshipsWho knows it and who owns it
ActivityUser interactions, document history and engagement patternsWhat is actually used and trusted

The activity pillar is the distinctive part. Glean does not only map which documents exist. It tracks which ones people open, edit and share, which helps it rank a heavily used design doc above a stale copy with the same title. Glean's connector docs describe the same flow: connectors "ingest, map, and normalize content, activity signals, and permission sets into a unified data layer" that feeds the Knowledge Graph (Glean connectors).

Evaluating enterprise AI for your team? See Coworker on your own tools in a 20-minute demo.

Book a demo

What is Glean's Enterprise Graph?

Glean announced the Enterprise Graph on September 25, 2025, alongside its third-generation Assistant (Glean blog). The Enterprise Graph sits on top of the Knowledge Graph and adds two things.

High-value entities. According to Glean's product page, the Enterprise Graph "builds on a rich knowledge graph that identifies high value entities" such as projects, people, customers and products, each connected to related signals such as the docs, support tickets and feature specs for a product (Glean Enterprise Graph). Glean's example is telling apart similar product names, such as two versions of the same model.

Personal graphs. Glean also builds a graph per employee. By "looking at actions across apps, these personal graphs can infer the broader tasks and projects behind them," which Glean uses to personalize Assistant and agents. Glean first wrote about this personal layer in June 2025 (Glean blog).

Glean says it builds these graphs "entirely using machine learning by understanding the data structures of enterprise apps and automatically inferring entities," and that this runs in each customer's single-tenant environment. Customers do not model the graph themselves.

How does Glean build its knowledge graph?

The build follows the connectors:

  1. Crawl. When a connector is first set up, Glean crawls content, the associated permissions and activity metadata (Glean docs).
  2. Index. Content and metadata go into a tenant-scoped search index used for retrieval, filtering and ranking.
  3. Map. Glean maps indexed content to people and activity to create the Knowledge Graph, which it calls a real-time model of the organization's indexed information.
  4. Enrich. The Enterprise Graph adds entities and personal graphs on top.

Glean is explicit that the graph is only as good as the connectors underneath it: building the Enterprise Graph "depends heavily on the quality of underlying connectors." Glean's site lists more than 275 connectors, far more than Coworker's 50+.

What does Glean use its knowledge graph for?

  • Search ranking and personalization. Results reflect who you are, who you work with and what is used, not only keyword or semantic match.
  • Glean Assistant. Answers are grounded in permission-aware content, with citations.
  • Glean Agents. Glean says the Enterprise Graph equips "employees and agents alike with multi-hop reasoning," and that Enterprise Graph actions let anyone build agents that use organizational context.
  • AI tools outside Glean. Through MCP, covered next.

Coworker

Answers from all your company knowledge

Coworker connects your tools and acts on what it finds, not just search.

Ask Coworker across your tools
Book a demo

Can you use Glean's knowledge graph in Claude, ChatGPT or Cursor?

Yes. Glean runs an MCP server so hosts such as Cursor, Claude and ChatGPT can "use Glean's permission-aware knowledge graph directly in those apps" (Glean MCP docs). Developers can also use Glean's REST APIs, which Glean recommends for "an integration that needs deterministic, structured results" (Glean developers).

Check the pricing model before you plan around this. Under Glean Enterprise Flex, client APIs, including the MCP server, consume FlexCredits "at current rate card" (Glean Enterprise Flex), on top of per-user seats. I compare the two MCP servers in Glean MCP vs Coworker MCP.

What are the limits of Glean's knowledge graph?

Glean is well reviewed: 4.7 out of 5 from 339 reviews on G2 in an August 2026 snapshot. G2's AI summary of those reviews lists the common complaints: AI answers that "can be inaccurate or hallucinated" (22 mentions), searches and answers that "can be slow or time out" (16), and pricing that users find high (G2).

Four structural limits are worth knowing:

  1. It is built for Glean's surfaces first. The graph powers Glean search, Assistant and agents. You can reach it over MCP and APIs, but it is priced and designed around Glean as the place people work. For how graph retrieval compares with plain embeddings in general, see knowledge graph vs vector database.
  2. Seats plus credits. Glean licenses per user per month and does not publish prices; advanced features and API or MCP use draw on FlexCredits. My Glean pricing breakdown collects the reported figures.
  3. The graph follows the index. Glean's model starts from indexed documents and their metadata, so what it knows about a tool depends on what that tool's connector captures and how often it syncs. Glean says as much: graph quality "depends heavily on the quality of underlying connectors."
  4. You shape the data, not the entity types. Glean lets you add custom metadata and properties to indexed content (Glean developers), but it describes the graph itself as inferred by machine learning. I did not find a documented way to define your own entity types for a specialized domain.

Glean Knowledge Graph vs Coworker OM2

Both are company knowledge graphs built for you from your work tools, both are permission-aware, and both are reachable over MCP. The differences are in how the graph is built, where it is meant to be used and what you can shape.

Glean Knowledge Graph and Enterprise GraphCoworker OM2
Built fromContent, people and activity across 275+ connectors (Glean's figure)Small, sourced facts from 50+ connectors
Unit of knowledgeIndexed documents mapped to people, activity and entitiesSmall, sourced facts linked to people, accounts and projects; outdated facts replaced and kept as history
Personal graphsYes, per employeeNot a stated feature
Custom domainsEntity types inferred by ML; custom metadata and properties supportedOM2 Custom Graph: ontology drafted from your data, or bring your own, approved by a person
Main surfaceGlean search, Assistant and agentsCoworker apps and agents, and Claude, ChatGPT and Cursor via Coworker MCP
MCP accessYes; client API and MCP use FlexCredits under Enterprise FlexYes
PricingPer user plus FlexCredits; not publishedNot published; book a demo

How OM2 builds its graph. OM2 connects to 50+ tools and keeps up as work happens, turns every message, doc and meeting into small, sourced facts linked to the people, accounts and projects they mention, marks a changed fact current while keeping the old one as history, and answers with exact graph queries within each person's access. In Coworker's words on its platform page, enterprise search returns documents to read, while OM2 pre-synthesizes facts, "so it answers relationships, not just lookups."

A note on numbers. Coworker's benchmark compares Claude through Coworker MCP with Claude using its own native connectors: with Coworker Learning on, 75.5% lower token cost across 100 tasks, about 9x lower on context-heavy retrieval tasks, and answers preferred 84.5% of the time in blind review, in Coworker's own statistically significant benchmarks comparing Coworker MCP with Claude's native tooling. It does not compare Coworker with Glean, and I have no head-to-head data, so I will not claim one is more accurate than the other.

Which should you choose?

Choose Glean if enterprise search is the main job, you need the widest connector coverage, you want personal graphs that tailor results to each employee, and your company is ready to adopt Glean as the place people search and ask questions. Its activity signals and years of search work are real strengths.

Choose Coworker if you want your company graph to work mainly inside the AI tools your teams already use, such as Claude, ChatGPT and Cursor, with agents that act on it; if you need to shape the graph for a specialized domain; or if you care about how many tokens each answer costs.

Look at both if you are comparing company knowledge graphs for AI agents. I cover the wider market in Glean alternatives, and compare the two products directly on Coworker vs Glean. Other matchups: Glean vs Claude, Glean vs Copilot and Rovo vs Glean.

Frequently asked questions

Does Glean use a knowledge graph?

Yes. Glean builds a Knowledge Graph from content, people and activity across its connectors, and uses it for search ranking, Glean Assistant and Glean agents. Since September 2025 it also builds an Enterprise Graph with business entities and personal graphs on top.

What is the difference between Glean's Knowledge Graph and Enterprise Graph?

The Knowledge Graph maps indexed content to people and activity. The Enterprise Graph, announced September 25, 2025, builds on it by identifying high-value entities such as projects, customers and products, and by adding personal graphs that infer each employee's tasks and projects.

How does Glean build its knowledge graph?

Glean's connectors crawl content, permissions and activity metadata from each app, index it in a tenant-scoped search index, and map it into the Knowledge Graph. Glean says it infers entities automatically with machine learning, in each customer's single-tenant environment.

Can I access Glean's knowledge graph from Claude or ChatGPT?

Yes, through Glean's MCP server, which works with hosts such as Claude, ChatGPT and Cursor and keeps Glean's permissions. Under Glean Enterprise Flex, MCP and client API use consumes FlexCredits.

Is Glean's knowledge graph permission-aware?

Yes. Glean's connectors ingest permission sets along with content and activity, and results are limited to what each user can already see in the source app.

What is a Glean alternative with a knowledge graph?

Coworker builds a company knowledge graph called OM2 across 50+ tools and serves it in Coworker's apps and in Claude, ChatGPT and Cursor via MCP. See my list of Glean alternatives for others.

Ready to get started?

Put Coworker to work inside your actual stack

Connect Salesforce, Slack, Jira, whatever you use, and run your first agent in minutes.