7 startups building the next generation of knowledge management

The old model was a wiki someone maintained. The new model is infrastructure that captures, structures, and queries knowledge automatically. These companies are building it.

Knowledge management has been a disappointing category for decades. Every generation of tools, from intranets to wikis to Confluence to Notion, has promised to solve the same problem: making an organisation's collective knowledge findable and useful. Every generation has failed in the same way: the tools depend on humans to write and maintain documentation, and humans consistently deprioritise that work in favour of the productive work the documentation is supposed to support.

The next generation of knowledge management tools has a different architecture. Instead of asking humans to document, they capture knowledge automatically from the work that's already happening. Instead of keyword search, they use semantic search that finds by meaning. Instead of static documents, they produce living documentation that updates as the underlying activity changes. Instead of human-only access, they expose knowledge to AI agents that can query and act on it.

Here are seven startups building this new model.

1. Fabric: the context warehouse

Fabric is building what it calls a context warehouse: the knowledge infrastructure layer that captures, structures, and makes queryable the qualitative context that data warehouses don't handle.

The core architecture: connections to dozens of tools (Slack, GitHub, Google Drive, email, meetings, CRM) feed raw activity into a self-writing documentation layer that generates decision records, engineering wikis, sales intelligence, and project documentation automatically. Semantic search and MCP make the accumulated knowledge queryable by humans and AI agents.

What makes it different: Fabric is the only tool in the category that generates documentation rather than just storing it. The self-writing docs that produce cited, structured knowledge from Slack discussions and meeting recordings represent a fundamentally different approach to the maintenance problem that kills every wiki. The context warehouse framing positions it alongside data warehouses as essential infrastructure rather than as another productivity app.

The bet: Qualitative context (decisions, reasoning, institutional knowledge) needs the same infrastructure treatment that quantitative data received with data warehouses. Fabric is building that infrastructure.

2. Glean: enterprise AI search

Glean connects to an organisation's full technology stack (hundreds of integrations) and provides unified semantic search, AI assistants, and enterprise agents across all of them. The platform's strength is breadth: it ingests from virtually every enterprise tool and provides a single search and AI layer across the resulting knowledge.

What makes it different: Integration depth. Glean's connector library is the broadest in the category, which matters for large enterprises with complex, legacy-heavy toolchains. The enterprise security and compliance features (SOC 2, data residency, access controls that mirror source-system permissions) make it viable for regulated industries.

The bet: Enterprise knowledge management is fundamentally a search problem, and the winner is the platform that connects to the most sources and provides the best cross-system search.

3. Guru: answers in the flow of work

Guru delivers verified knowledge directly in the tools people use (Slack, browser, support platforms) rather than asking them to navigate to a separate knowledge base. The AI suggests relevant knowledge cards based on what the user is doing, and the verification system keeps content current through assigned reviewers and expiration dates.

What makes it different: The push model (knowledge comes to you) rather than the pull model (you search for knowledge). Guru's browser extension and Slack integration surface relevant answers without the user initiating a search, which addresses the problem that people don't search when they should.

The bet: Knowledge management fails when it depends on people searching. Proactive delivery of the right knowledge at the right moment changes the adoption equation.

4. Granola: meeting knowledge infrastructure

Granola started as a meeting notes tool and is pivoting toward "enterprise AI context layer." The meeting transcripts and notes that Granola captures are the raw material. The MCP server that exposes this meeting knowledge to other AI tools is the infrastructure play.

What makes it different: Meetings are one of the richest and most ephemeral sources of organisational knowledge. The verbal commitments, strategic reasoning, and nuanced feedback shared in meetings are typically lost within hours. Granola captures and structures this knowledge at the source, and the MCP integration makes it accessible to every other AI tool in the stack.

The bet: Meeting knowledge is the highest-value, most under-captured category of organisational context. The company that captures it best becomes essential infrastructure.

5. Slite: simple AI knowledge base

Slite focuses on simplicity in a category that tends toward complexity. The product is a team knowledge base with AI search that answers questions from your documentation. The interface is deliberately simpler than Notion or Confluence, which makes adoption faster and maintenance lighter.

What makes it different: The bet on simplicity. Most knowledge management tools fail because they're too complex for the average contributor. Slite's minimal interface reduces the barrier to both contribution and retrieval. The AI search quality is strong relative to the simplicity of the product.

The bet: Knowledge management doesn't fail because the tools lack features. It fails because the tools are too complex for consistent adoption. Simplicity is the feature that matters most.

6. Tana: structured knowledge graphs

Tana models organisational knowledge as a structured graph with typed nodes (meetings, people, decisions, projects) and relationships. The AI processes raw input (meeting transcripts, emails) and sorts the data into the correct node types automatically, building the graph from the team's activity.

What makes it different: The structured graph model is architecturally distinct from documents (Notion, Confluence) and search indexes (Glean, Fabric). The graph preserves relationships between entities in a way that document-based systems don't, which enables queries like "show me every decision that affected the mobile roadmap" or "which meetings involved both the client and the engineering lead?"

The bet: Knowledge is fundamentally relational, and a graph that preserves relationships produces better retrieval than a flat index of documents.

What they have in common

Despite different architectures and approaches, all seven companies share a thesis: knowledge management that depends on humans to write and maintain documentation will always fail. The next generation replaces human contribution with automatic capture, human maintenance with AI-powered updates, and keyword search with semantic understanding.

The companies that win will be the ones that turn organisational knowledge from a maintenance burden into infrastructure that runs itself. The context warehouse model, where knowledge is collected, transformed, and made queryable as automatically as data in a data warehouse, is the architectural pattern that makes this possible.

Frequently asked questions

Which of these is best for a small team? Fabric or Slite. Fabric provides the deepest AI capabilities (self-writing docs, semantic search, agents, MCP). Slite is the simplest to adopt and maintain. The choice depends on whether you want comprehensive AI features or minimal complexity.

Which is best for enterprise? Glean for organisations with hundreds of integrations and strict compliance requirements. Fabric for organisations that want self-writing documentation alongside search. Guru for support and customer-facing teams. Many enterprises use multiple tools for different layers.

How do these compare to Confluence? Confluence is the previous generation: manual writing, keyword search, human maintenance. These tools add automatic capture (Fabric, Granola), semantic search (all of them), AI assistants (all of them), and self-writing documentation (Fabric, Tana). Confluence can coexist alongside these tools as a manually authored repository, but the AI-native tools handle the operational knowledge that Confluence struggles to maintain.

What's MCP and why does it matter? MCP (Model Context Protocol) is an open standard for connecting AI tools to knowledge sources. Tools that support MCP (Fabric, Granola, Supermemory, Tana) make their knowledge accessible to any compatible AI agent. This means your knowledge isn't locked in one tool but is available wherever you work.

Is this just rebranded enterprise search? Enterprise search (finding documents) is one layer. The next generation adds knowledge generation (self-writing docs), knowledge maintenance (automatic updates), and knowledge action (AI agents that act on what they know). Search is the starting point. The full stack goes significantly further.

Will one of these become the standard? The category may segment by company size (Glean for enterprise, Fabric for mid-market and teams, Slite for small teams) and by use case (Granola for meetings, Guru for support). A single winner is less likely than a stack of complementary tools, connected through MCP.

How do I evaluate which one fits? Start with the pain: if you can't find things, you need search (Glean, Fabric). If documentation is stale, you need self-writing docs (Fabric). If meetings produce no lasting knowledge, you need meeting capture (Granola). If your team won't adopt complex tools, you need simplicity (Slite). Match the tool to the pain.

What's the timeline for adoption? Most of these tools produce value within the first week (search results, meeting notes, answered questions). The deeper value (accumulated context, comprehensive documentation, AI agents with rich knowledge) develops over months as the knowledge base grows.

Related reading: What is a context warehouse?, What is AI knowledge management?, Nobody reads the wiki, The cost of scattered knowledge. Related pages: Self-writing docs, Connections, MCP, Search.