Knowledge Graph Overview
A persistent memory layer that stores entities, relationships, and insights across all workflow executions. Future runs receive compact, relevant context — not a dump of raw history.
What is the Knowledge Graph?
The Knowledge Graph (KG) is AgentLed's long-term memory system. Unlike traditional automation tools that start from zero every run, AgentLed stores entities (companies, people, deals, leads), their relationships, and AI-generated insights in a graph database.
Every workflow execution can read from and write to the graph. A lead scoring workflow can reference outcomes from previous runs. An investor matching workflow can factor in historical decisions and partner corrections. A content workflow can learn from engagement patterns.
The key mechanism is not injecting full history into prompts — it is building compact, structured profiles per entity that carry forward only what's relevant.
How Memory Compounds: Score → Store → Summarize → Retrieve
The Summarize step is what keeps prompts bounded. Raw scoring logs are never dumped into future runs — only the compact profile that distils what matters.
Architecture
The Knowledge Graph has three layers:
Entity Nodes
Companies, investors, leads, contacts, deals — any business object. Each node has a type, properties, and a unique identity. Nodes are created automatically during workflow execution or imported from Knowledge Lists.
Relationship Edges
Typed connections between entities: SCORED, APPROVED, REJECTED, CONTACTED, MATCHED. Edges carry metadata — a SCORED edge includes the score value, rationale, and the execution that created it. Query edges to traverse the graph or build compact profiles.
AI Insights
Insights generated by AI steps: scoring rationale, pattern recognition, calibration data, partner corrections. Insights are linked to the entities and executions that produced them, creating an audit trail of AI reasoning over time.
Built-in KG Tools
Every AI step in a workflow automatically has access to four Knowledge Graph tools. No configuration needed.
| Tool | Description |
|---|---|
| kg_search | Natural language semantic search across all entities and insights |
| kg_traverse | Follow typed relationship edges from a starting entity |
| kg_nodes | List all entities of a given type (e.g., all investors, all leads) |
| kg_write | Persist AI-generated insights and scores back to the graph |
What Changes Between Runs
The difference between a cold-start run and a context-aware run is the compact profile. Here are two examples — one investor matching workflow, one lead scoring workflow — showing what the AI receives at Run 1 versus Run 5.
Investor matching — Benchmark Capital
Run 1 — Cold Start
Context available:
Sector focus: Enterprise SaaS, DevTools Stage: Series A–B Check size: $5M–$25M Portfolio: 34 companies (public data)
Score
74 — “Strong sector fit, check size aligned. No prior relationship data.”
Run 5 — With Compact Profile
Context available:
Sector focus: Enterprise SaaS, DevTools Stage: Series A–B Check size: $5M–$25M — Compact profile (4 prior runs) — Approved for intro: 2 SaaS deals (both converted to meetings). Passed: 1 marketplace deal. Partner note: "Strong PLG preference. Weak fit on sales-led GTM." Last outcome: Committed at lead round.
Score
88 — “Confirmed SaaS fit with PLG signal. Prior approval pattern and commitment outcome support proceed.”
Lead scoring — Jane Smith
Run 1 — Cold Start
Context available:
Title: Head of Marketing Company: Acme Corp (180 employees, SaaS) Industry: B2B SaaS LinkedIn: Active (3 posts/month)
Score
65 — “ICP match on size and industry. No prior contact data.”
Run 5 — With Compact Profile
Context available:
Title: Head of Marketing Company: Acme Corp (180 employees, SaaS) Industry: B2B SaaS — Compact profile (4 prior runs) — 2 emails sent. Opened 2 of 2. No reply. Colleague (CTO, Mark T.) was qualified last quarter — meeting booked. Rep note: "Likely decision-influencer, not buyer. Needs CTO intro first."
Score
71 — “Re-engagement candidate. Warm path via CTO colleague. Route to account-based sequence.”
The compact profile is built from structured decisions stored in the KG — not from replaying raw execution logs. Prompt size stays bounded regardless of how many prior runs exist.
Data Sources
Data enters the Knowledge Graph through three paths:
- 1.Knowledge Lists — Upload CSVs or create structured lists in the dashboard. Rows sync to the graph as entities.
- 2.Workflow execution — AI steps write scores, rationale, corrections, and outcomes as they process data.
- 3.API ingestion — Push historical data or external enrichment results via the ingestion endpoints.
Next Steps
- Scoring & Feedback — Full write/read path and compact profile details
- KG API — Query and write to the graph programmatically
- Business Memory — Workspace-level shared context
- Knowledge Lists — Structured data import and storage