Ensue: The Shared Memory Layer for Agents

Instead of one siloed, stateless agent or LLM, imagine a world where you have access to the shared memory of millions - all connected through a single, cohesive system.

Introducing Ensue

The first shared memory layer for agents. It will be a distributed protocol and memory marketplace where

• Memory is portable across tools, agents, and models.
• Agent builders tap into rich, shared memory streams from other actors.
• Triggers and access rules for the memory snippets (per-agent, per-field) will shape and secure interactions.

How Ensue will change your agent-building

• Build smarter systems powered by collective intelligence.
• Create dynamic, programmatic workflows that reason across shared memory.
• Move beyond rigid, pre-scripted tasks toward adaptive, autonomous coordination at scale.

Problems Ensue solves

• Scaling multi-agent coordination – Tired of message-passing hacks and database boilerplate? Tools like A2A are great - until you scale. Ensue gives agents a shared memory space to collaborate.
• Porting context across tools – Pick up a conversation in Claude Code that started in ChatGPT or Gemini. Or pass context from an n8n workflow into another custom agent.
• Controlling what agents can access – With agent permissions, you can give each agent exactly what they need - and nothing more.

Ensue in action

Imagine a multi-actor system of specialized agents and users (app-users, legal-counsel, feature-developer, etc.) using unique models (Gemini, Claude, GPT, etc.) dynamically interacting with the same Ensue shared memory layer.

import { AgentMemory, Fq } from "ensue";

// Gemini powered PM agent that takes user inputs and develops feature specs
async function main() {
  const geminiKey = Fq.random();
  // loads existing memory from `address` config file
  const memory = await AgentMemory.create(geminiKey);
  const codeBase = await memory.read("code-base");

  // user agents constantly append feedback
  const userAppFeedback = await memory.read("user-app-feedback");

  // ensure critical data is authenticated
  const managerAppFeedback = await memory.readVerified("manager-app-feedback");
  const lawyerAppFeedback = await memory.readVerified("lawyer-app-feedback");

  const memories = [
    codeBase,
    userAppFeedback,
    managerAppFeedback,
    lawyerAppFeedback,
    // prompt:
    "write me a feature spec that addresses all the feedback above",
  ];

  const instructions = await executeGemini(memories);
  await memory.update_or_create_key("gemini-instructions", instructions);
}

async function main() {
  const claudeKey = Fq.random();
  // loads same memory from `address` config file
  const memory = await AgentMemory.create(claudeKey);
  const featureSpec = await memory.read("gemini-instructions");
  const newFeature = claudeCodeProcess(featureSpec);
  claudeMemory.update_or_create_key("new-feature", newFeature);
}

Get involved 🤝

Private alpha starts next week. We’re letting in builders gradually. Get on the list early:
→ Sign up for early access
→ Ask questions, share feedback. @ensue_ai on X, our DMs are open.