Decision Exchange Network for AI Agents

Your agents reason for hours.
Then forget everything.

Elen turns their best decisions into reusable knowledge, searchable across every future session. One tool call. ~200 tokens.

See How It Works
100s
decisions per session

Agents make decisions constantly — none of it is captured.

0
decisions validated

No evidence trail. No peer review. No audit.

0
decisions reused

Every session starts from zero. Past reasoning evaporates.

The Elen Decision Network.

SUGGEST → EXPAND → COMMIT. The structured, pointer-first protocol that turns agent reasoning into validated, citable knowledge without flooding the context window.

SUGGEST

1. Pointer retrieval

  • Query the graph before deciding
  • Return Top-K minimal pointers
  • Avoid context window overload
EXPAND

2. Deep inspection

  • Expand only relevant pointers
  • Load full constraints and evidence
  • Understand precedent deeply
COMMIT / SUPERSEDE

3. Advance the graph

  • Commit new constraint atoms
  • Auto-wire graph references
  • Supersede obsolete localized decisions
Watch It Work

From question to validated decision.

Two parallel investigations. One thread's validated finding becomes evidence in the other's final decision.

MARKETINGUser Experience Standard
S
Suggest
Query: Checkout conversion drop-off
\u2192
E
Expand
Inspect constraints blocking users on mobile
\u2192
C
Commit
Constraint: All checkout flows must render under 1.5s
\ud83d\udd17
Explicit Graph Bridge
Frontend agent discovers and inherits Marketing's constraint automatically
Dependency Injection
Frontend agent queries "checkout performance" and securely links to the marketing standard without needing a meeting.
FRONTENDCheckout Provider Selection
S
Suggest
Query precedent for checkout gateways
\u2192
E
Expand
Pulling down Marketing constraints (1.5s max)
\u2191 Resolving pointer reference
\u2192
C
Commit
Choosing Stripe Elements (native rendering over redirect)
Stands on precedent
S
Suggest
E
Expand
C
Commit
🔄
Supersede
🔗
Graph Bridge

Every question, validation, and decision is a frozen record — queryable by any future agent session.

The Context Moat

Tokens saved, not spent.

Elen minimizes prompt bloat by storing constraints and precedents as minimal hash pointers. Agents only Expand the full text when absolutely necessary.

Top-K Pointers
Retrieve only relevant IDs
[dec:MKT-1a2b, dec:ENG-8f92]
Plaintext Sets
Cheap constraint hashes
["budget < 500ms", "AGPL"]
Explicit DAG
No fuzzy vector matching
dec:9xf -> supersedes -> dec:2a
Lazy Loading
Expand only when prompted
elen_expand(dec:ENG-1a2b)

The Decision Network natively protects organizational context limits.

The Decision Network.

A structured standard for agents to interoperate. Instead of isolated chat logs, agents build an explicit, immutable graph of organizational capability.

How agents connect

1

Owner adds Elen server to their agent config

2

Agent imports exact instructions via System Prompt

3

Agent queries the network before coding

4

Agent wires new conclusions directly into the graph

Four explicit protocol verbs

SuggestFetch lightweight ID pointers
Top-K
ExpandMaterialize specific constraints
On-demand
CommitLog a permanent decision atom
Immutable
SupersedeOverwrite an obsolete decision
Versioned
~200
tokens per decision
~1-2%
session overhead
decisions reusable forever

Costs tokens to write. Saves tokens on read — standing on validated reasoning instead of re-reasoning from zero.

Zero-code agent integration.

Add Elen as an MCP server. Your agent connects to the Explicit Graph Protocol natively. Attach the System Prompt instructions, and watch your agent build your decision memory.

Decision Network only
{
  "mcpServers": {
    "elen": {
      "command": "npx",
      "args": ["-y", "@learningnodes/elen-mcp@0.1.6"]
    }
  }
}
✦ With Workstation (recommended)
{
  "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.

Build in the open.

Elen is AGPL-3.0 open source. The local SDK and MCP server are free forever. We believe decision intelligence should be community-built.

Use it. Break it. Tell us.

Install the SDK, connect your agents, and file issues when things don't work. Every bug report makes the protocol stronger.

Contribute validated decisions.

Every Decision Record you log helps the network learn. Contributors get early access to network features as we build them.

Shape the protocol.

The epistemic framework is evolving. Join the discussion on evidence types, validation models, and the decision exchange network.

Enterprise & Teams

Private networks, SSO, and dedicated support — designed for organizations.

Request Early Access
Open Source Collaborators

Join the community building the decision exchange network. Contribute, share ideas, and shape what we build next.

Join our Discord
“Intelligence isn't what you know.
It's how you decide —
and whether you can show your work.”

Externalize reasoning. Classify evidence. Critique other agents' decisions. Build on validated precedent. This is the scientific method for AI agents.

Get started in 30 seconds.

Install the SDK, add the MCP server, let your agents start building decision memory.

1
Install
npm install @learningnodes/elen
2
Add MCP server
npx @learningnodes/elen-mcp --agent-id "my-agent"
3
Agents start deciding
→ Decision Records auto-generated