Knowledge Graph
AgentLed's Knowledge Graph is a structured memory layer that stores entities, relationships, scoring history, and workflow outcomes across all runs in a workspace.
It is not just a database and not just prompt memory. It is a graph of business knowledge that workflows and agents can read from and write to over time, so future decisions use prior context instead of starting from zero.
Nodes
companies, people, deals
Edges
scored, matched, contacted
History
scores, rationale, outcomes
Access
workflows, agents, MCP
Why it matters
When a customer asks, “Have we seen this investor before?” or “What happened with similar startups last time?”, the answer comes from structured memory, not guesswork.
1. What is it?
The Knowledge Graph stores the objects your business cares about, how they relate to one another, and what your workflows learned about them. That memory compounds over time.
Companies, investors, startups, contacts, deals, and any other business object your workflows work with.
Typed edges like SCORED, CONTACTED, MATCHED, or INVESTED_IN that connect entities across runs.
Past scores, rationales, and outcomes that let future runs calibrate decisions instead of starting cold.
Tenant-isolated context shared across workflows, agents, and MCP/API access in the same workspace.
2. Workflows
The workflow layer is where the graph becomes useful. Runs do not only produce output — they also update shared memory for the next run.
Write
Execution results become durable context: enriched contacts, scored leads, matched investors, discovered relationships, and final outcomes.
Read
A workflow can ask: have we scored this entity before, what happened last time, which relationships already exist, and what was learned from previous outcomes?
Compound learning
Instead of repeating the same analysis from scratch, each run builds on what earlier runs already discovered and validated.
Workflow run
Enrich startup, score investor, or qualify a lead.
Write to KG
Store entities, scores, relationships, rationale, and outcomes.
Next run reads
Use prior context before making the next decision.
Typical write examples
Typical read questions
3. Agents
Agents use the graph as persistent business context. That matters in both chat and autonomous execution: the agent can retrieve past decisions, store new findings, and share them with the rest of the workspace.
Agents can read workspace memory while chatting with users or completing autonomous tasks.
Agent memory is backed by the Knowledge Graph, so decisions and findings do not disappear when a session ends.
One agent can store an evaluation and another agent can reuse it later without redoing the same analysis.
Research, sourcing, matching, and outreach agents all operate on the same business memory layer.
Multi-agent example
Sourcing agent
Evaluates a startup and stores thesis fit, sector notes, and scoring rationale in the KG.
Matching agent
Reads that evaluation later when looking for investors, so it starts with known context instead of redoing the same research.
This is the practical difference between separate agents and coordinated agents: one agent's findings become available to all agents in the workspace through the same memory layer.
4. Examples
The point of the graph is not abstract “memory.” The point is that repeated work becomes more informed, more coordinated, and less wasteful.
Week 1
Enrichment writes to the graph
Week 2
Outreach reads before acting
Why it matters: Work does not fragment between runs because the workflow state lives in shared memory.
Run 1
Static profile only
Run 8
Score uses actual history
Why it matters: The value is not one score. The value is a score that gets better as real outcomes accumulate.
First batch
Initial screening
Second batch
Screening with calibration context
Why it matters: The KG turns repeated sourcing into a learning loop instead of a repeated manual exercise.
5. Comparison
n8n, Zapier, and Make can orchestrate steps. The Knowledge Graph adds durable business context across executions.
| Capability | n8n | Zapier | Make | AgentLed KG |
|---|---|---|---|---|
| Shared memory across runs | No | No | No | Yes — persistent business memory |
| Relationship graph between entities | Manual | Manual | Manual | Built in |
| Scoring history and outcome tracking | Not native | Not native | Not native | Built in |
| Cross-workflow context reuse | Limited | Limited | Limited | Shared across workspace |
| Agents can query memory directly | No | No | No | Yes — via MCP and workflow tools |
| Compound learning from outcomes | Custom build | Custom build | Custom build | Native pattern |
6. Technical access
The graph is accessible through MCP tools, AgentLed's API surface, and the CLI/MCP server. That makes it usable both inside workflows and from external technical tooling.
MCP tools
query_kg_edges
Traverse relationships for an entity and inspect how companies, people, deals, or scores connect.
get_knowledge_rows
Read structured rows from a knowledge list that workflows store and update across runs.
get_scoring_history
Fetch past scoring decisions and outcomes to calibrate the next decision.
get_knowledge_text
Retrieve text-based knowledge such as ICP notes, investment theses, or operating rules.
Programmatic access
Share this page with a technical contact, then show them the workflow. The Knowledge Graph is the layer that makes each next run more informed than the last.
