Knowledge Graph
Your automation shouldn't start from scratch every time. AgentLed's Knowledge Graph gives your workflows persistent, compounding memory — they get smarter with every execution.
Every organisation carries institutional knowledge: what "qualified" means to your sales team, which investors match your fund's thesis, which content formats convert best for each client.
Today that context lives in people's heads, buried in Slack threads and docs nobody reads. Your AI agents never see it — so they guess. And they guess differently every time.
You
"Is Acme Ventures a good fit for our Series A?"
Sales Agent · no shared context
"They invest in B2B SaaS, probably a good match."
Research Agent · no shared context
"Their last fund was $200M, might be too large for us."
Different agents, different answers. No shared memory = inconsistent decisions.
Every execution stores structured results — leads, scores, insights, outcomes. No manual data entry. Your KG grows as you work.
Previous scores, historical patterns, learned preferences. New runs build on old ones instead of starting from zero.
One workflow learns it, every workflow benefits. Your sales agent and research agent see the same truth.
Score an investor. Track the IC outcome. Feed it back. Next scoring run is more accurate. This is compound learning.
Data sources feed in. Workflows read and write. Knowledge compounds. Every agent stays aligned.
Data Sources
Knowledge Graph
entities + relationships + scores
Workflow Agents
Every workflow execution makes the next one smarter. Not just storing data — learning from outcomes.
New Entity
"Series A Fintech" pattern added to scoring model
Updated Concept
"Qualified Investor" definition refined from IC outcomes
Prediction Validated
DMF score 8.2 → investor committed (89% accuracy)
Feedback Loop
Sector weight adjusted: fintech +12% based on 23 deals
prediction accuracy after 12 scoring runs. Started at 62%.
investor profiles enriched and scored with compounding context.
manual data entry. Workflows write to KG automatically.
It's not just retrieval. It's shared understanding.
| RAG | AgentLed KG | |
|---|---|---|
| What it retrieves | Chunks of text by similarity | Structured entities, relationships, and scores |
| Company terminology | Doesn't know your internal terms | Learns and stores your definitions |
| Role-specific meaning | Can't distinguish context per workflow | Meaning scoped per agent and workflow |
| Learning | Static — doesn't update from activity | Evolves continuously as workflows run |
| Cross-workflow sharing | Each query retrieves independently | One workflow learns it, every workflow knows it |
| Prediction tracking | Not possible | Score → track outcome → improve next run |
Scores compound across 3,000+ profiles. Each IC meeting outcome feeds back to improve future scoring. Prediction accuracy went from 62% to 89% over 12 runs.
ICP scores improve as you learn which leads actually convert. The KG tracks what happened after outreach — closed, ghosted, wrong fit — and adjusts scoring weights.
Track which topics, formats, and channels perform best. Feed performance data back into content generation. Each publishing cycle gets more targeted.
Each client workspace builds its own knowledge layer. SEO audit findings, keyword performance, content history — all compound over time under your white-label brand.
MCP Tools
query_kg_edges — traverse entity relationships
get_knowledge_rows — retrieve structured data
get_knowledge_text — access text content
get_scoring_history — prediction vs outcome tracking
list_knowledge_lists — browse all knowledge bases
Infrastructure
Built on DynamoDB with caching layer
Tenant-isolated (GDPR compliant)
Graph edges model entity relationships
Sub-100ms query response time
Automatic schema evolution as agents work
Every execution makes the next one smarter. Start compounding.
