Elen — Decision Network Walkthrough

1 Suggest (Pointer Retrieval)

Before making a decision, the Agent asks Elen if there is precedent. Elen returns minimal pointers to avoid flooding the context window.

AI Agent Encountering Problem

The agent is building a high-traffic API and needs to select the correct caching layer.

AI Agent Queries Precedent

TOOL CALL: elen_suggest { "query": "caching database architecture", "domain": "infrastructure" }

Elen MCP Returns Minimal Pointers

Elen searches the graph and returns just IDs and summaries. The agent's context window stays light.

{ "candidates": [ { "decision_id": "dec:INF-a6b8c9", "decision_text": "Use Redis for ephemeral high-throughput caching (1k+ RPS)", "constraint_set_id": "cs:7f8d2b", "status": "active" } ] }

2 Expand (Deep Inspection)

The Agent sees a relevant pointer and explicitly requests the full node details to load constraints and evidence.

AI Agent Requests Expansion

TOOL CALL: elen_expand { "decisionId": "dec:INF-a6b8c9" }

Elen MCP Rehydrates Atom

Elen pulls the full Decision Record, mapping the exact constraints the previous agent locked in.

{ "id": "dec:INF-a6b8c9", "text": "Redis chosen over Memcached due to data-structure support.", "constraints": [ "Must support pub/sub capabilities", "Requires high availability via Sentinel", "Latency < 2ms p99" ] }

3 Commit (Advance Graph)

The Agent decides to use Redis based on the precedent, but for a new specific microservice. It commits the new localized node.

AI Agent Logs New Decision

TOOL CALL: elen_commit { "question": "Caching layer for the User Auth Microservice", "domain": "auth", "decisionText": "Using Redis, citing previous INF precedent.", "constraints": ["Auth tokens must TTL natively"], "refs": ["dec:INF-a6b8c9"] }

Graph Network Immutable Write

Elen hashes the constraints, links the reference, and saves the new atom into the network.

✅ Validated graph reference to dec:INF-a6b8c9

✅ Constraint Set Hash created: cs:7f8d2b...

✅ Atom immutable freeze sequence applied

dec:INF-a6b8c9

Redis core tech choice

Active Precedent

dec:AUT-b2c4d6 (NEW)

User Auth Redis Layer

Active Leaf

4 Supersede (Fixing Mistakes)

Six months later, an Agent notices Redis is blowing out memory on the Auth service due to unoptimized TTLs. It proposes a change.

AI Agent Revokes Past Node

Instead of just committing a disconnected node, the Agent explicitly flags the old node as obsolete.

TOOL CALL: elen_supersede { "oldDecisionId": "dec:AUT-b2c4d6", "question": "Caching layer for the User Auth Microservice", "domain": "auth", "decisionText": "Replacing Redis with DynamoDB DAX for auth due to OOM errors under load.", "constraints": ["Must prevent node memory fragmentation"] }

Graph Network State Transition

Elen creates the new decision and deprecates the old one. Future queries for "auth" will only return the DynamoDB precedent.

✅ Existing dec:AUT-b2c4d6 marked SUPERSEDED

✅ New dec:AUT-f9e8d7 marked ACTIVE

dec:AUT-b2c4d6

User Auth Redis Layer

Superseded

dec:AUT-f9e8d7 (NEW)

User Auth Dynamo DAX

Active Leaf

The Network Guarantees

🔍

Context Light

~200

Base tokens per query cycle.

📂

On Demand

Pointers

Bloated data only injected when manually expanded.

🕸️

Traceable

Graph

Decisions are explicitly wired via refs arrays.

⚡

Self Healing

Supersedes

Agents debug past agents and formalize the fix.

Get Started in 60 Seconds

Add Elen as an MCP server in your IDE. No SDK required.

Decision Network only

Suggest, Commit, Expand, Supersede. Data stays on your machine.

{
  "mcpServers": {
    "elen": {
      "command": "npx",
      "args": ["-y", "@learningnodes/elen-mcp@0.1.6"]
    }
  }
}

✦ With Workstation recommended

Auto-connects app.elen.learningnodes.com to your local decisions.

{
  "mcpServers": {
    "elen": {
      "command": "npx",
      "args": ["-y", "@learningnodes/elen-mcp@0.1.6"],
      "env": {
        "ELEN_LOCAL_API": "true"
      }
    }
  }
}

Works with Claude Code, Cursor, Windsurf, VS Code Copilot, and any MCP-compatible agent.

Elen Decision Network