Knowledge Graph

The Knowledge Graph is FanDesk's institutional memory engine — a continuously updated graph of every entity, relationship, and piece of context across your entire workspace. It's what makes FanDesk's AI genuinely intelligent rather than just pattern-matching on recent context.

The core insight: when all your data lives in one place — chat, tasks, email, documents, meetings, contacts — you can build a graph that connects it all. That graph becomes your organization's persistent memory.

What Is a Knowledge Graph?

A knowledge graph is a structured representation of entities and the relationships between them. Instead of storing text in isolated buckets, FanDesk extracts structured knowledge and stores it as connected nodes.

Entity example:

• "Acme Corp" is a Company entity
• Connected to: John Smith (contact), the Mobile App project (project), 3 email threads, 2 meeting notes, the Q3 proposal (page)
• Tagged with: "enterprise client", "healthcare", "renewal in Q4"

When you ask "What's our status with Acme Corp?", DeskMate traverses this graph to give you a complete, cross-module answer — not just a keyword search result.

How the Pipeline Works

1. Capture (Real-Time Indexing)

Every piece of content your team creates is automatically indexed as it's created or updated:

• Chat messages and threads
• Tasks, comments, and status changes
• Pages and document edits
• Emails (sent and received)
• Meeting transcripts and summaries
• Contact notes and interactions
• File names and metadata

No manual tagging. No configuration. It happens automatically.

2. Embed (Vector Representation)

Each piece of content is converted into a vector embedding using Voyage AI (1024-dimensional vectors). This powers semantic search — finding content by meaning, not just matching exact words.

Embeddings refresh every 10 minutes, so new content becomes searchable almost immediately.

3. Extract (Entity Recognition)

An AI extraction pipeline runs every 15 minutes, scanning all new content to extract:

4. Connect (Relationship Mapping)

Extracted entities are linked based on context:

• "Sarah is leading the API redesign" → Sarah → works on → API Redesign project
• "Acme Corp signed the mobile contract" → Acme Corp → client for → Mobile App project
• "The outage was caused by the deployment pipeline" → Outage incident → related to → Deployment Pipeline concept

5. Maintain (Deduplication & Cleanup)

The graph maintenance pipeline (daily at 2:30 AM):

• Deduplicates entities — "Acme Corp", "Acme", and "Acme Corporation" are merged into one node
• Strengthens active connections — Relationships mentioned repeatedly become stronger signals
• Fades stale connections — Old relationships that haven't been reinforced lose weight over time
• Merges duplicate nodes — Different representations of the same entity are unified

Entity Merging and Deduplication

One of the graph's most powerful features: it knows that "John", "John Smith", and "@john" in different contexts all refer to the same person.

When FanDesk detects duplicate entities:

1. An automated merge is proposed based on name similarity and co-occurrence patterns
2. The merge is applied: all edges from both nodes are transferred to the surviving node
3. You can manually trigger a merge from the Knowledge Graph view if you spot duplicates

This keeps the graph clean and prevents fragmentation — a common problem in knowledge systems.

What the Knowledge Graph Powers

DeskMate's Intelligence

Before DeskMate responds to any query, it enriches its context with a knowledge graph lookup:

• Searches for entities mentioned in your query
• Traverses relationships to find connected context
• Surfaces relevant people, projects, emails, tasks, and pages — even ones you didn't mention
• The result: DeskMate "knows" your organization without you explaining it every time

Example: Ask "What's the status of the Acme deal?" — DeskMate traverses from the Acme Corp entity to find all connected tasks, emails, contacts, and meeting notes, then gives you a synthesized answer across all of them.

Semantic Search Across All Modules

Universal search that understands meaning:

• "Find everything about the pricing redesign" — returns tasks, messages, emails, pages, and meetings, ranked by relevance
• "Who worked on authentication?" — finds people connected to auth-related entities, not just keyword matches
• "What decisions did we make about the mobile app?" — extracts decision entities and returns the pages and meetings where they were documented

Daily Digest

Your morning briefing is powered by the knowledge graph — it surfaces connections and updates most relevant to your current work based on entities you're connected to.

Context Enrichment for AI Employees

AI employees (like Riyan, the External Relations AI) use the knowledge graph as their long-term memory. The more your team uses FanDesk, the more context the AI employees have — making them genuinely more effective over time.

Knowledge Graph in Practice

Scenario: Preparing for a Client Meeting

Before a call with Acme Corp:

1. Ask DeskMate: "Brief me on Acme Corp before the 3pm call"
2. DeskMate traverses the Acme Corp entity in the knowledge graph
3. It surfaces: last 5 emails with Acme, open tasks related to their project, the Q3 proposal page, your last 2 meeting summaries with them, and any recent Slack messages mentioning Acme
4. You get a comprehensive brief in seconds — not 20 minutes of searching across tools

Scenario: Onboarding a New Team Member

When a new engineer joins:

1. They start working on tasks, reading pages, and messaging teammates
2. The knowledge graph learns their areas of work (from task assignments and page edits)
3. When someone asks DeskMate "Who knows about the payment integration?", the graph routes to the right person — even if they're new
4. The new hire immediately has access to all institutional context via DeskMate, not just what was manually documented

Scenario: Post-Incident Analysis

After a production incident:

1. The incident is documented in FanDesk Incidents
2. The knowledge graph extracts: affected systems, people involved, timeline, root cause
3. When a similar issue occurs months later, DeskMate can surface the previous incident: "We had a similar issue on March 15th — see the postmortem"
4. Institutional memory persists even after the engineers who handled the incident have left the company

Why This Matters: The Compounding Advantage

Most SaaS tools have flat data — your knowledge is scattered across disconnected databases. When people leave, their context leaves too. When you switch tools, you lose everything.

The FanDesk knowledge graph compounds:

• Day 1: Basic entity extraction from initial content
• Month 1: Rich relationship map of your projects, clients, and team
• Month 6: Deep institutional memory — past decisions, client history, technical debt, relationship context
• Year 1: Organizational intelligence that persists through team changes and is impossible to replicate in a new tool

The switching cost isn't the interface — it's the accumulated knowledge you'd lose. That's the moat.

Background Processing (No Setup Required)

The knowledge graph is fully automated:

• Vector embeddings refresh every 10 minutes
• Entity extraction pipeline runs every 15 minutes
• Graph maintenance (deduplication, cleanup) runs daily at 2:30 AM
• No manual tagging, no configuration, no maintenance needed

Everything happens in the background. You just use FanDesk normally — the knowledge graph builds itself.

---

Next: Learn about the people directory in People Directory.