Hive Mode - Cross-Application Memory

Info

Last Updated: 2026-01-08

Share memory across multiple agents and applications.

Overview

Hive Mode enables multiple agents/tools to share a single memory space, creating a unified context across your entire AI ecosystem.

Setup

const cortexMemory = createCortexMemory({
  convexUrl: process.env.CONVEX_URL!,
  memorySpaceId: "shared-workspace", // Same across apps
  userId: currentUser.id,
  agentId: "web-assistant", // REQUIRED: identifies the agent in user-agent conversations
  hiveMode: {
    participantId: "web-assistant", // Optional: tracks which tool/agent stored data
  },
});

Understanding agentId vs participantId

• agentId (required): Identifies the agent participating in a user-agent conversation. This is the agent that responds to the user.
• participantId (optional, hive mode): Tracks which tool/agent stored a particular piece of data within a shared memory space. Used for attribution and filtering.

Use Cases

1. Next.js App + Cursor + Claude

// Next.js App
const webAppMemory = createCortexMemory({
  convexUrl: process.env.CONVEX_URL!,
  memorySpaceId: "shared-workspace", // Same across all three
  userId: currentUser.id,
  agentId: "web-app-agent",
  hiveMode: { participantId: "web-app" },
});

// Cursor (if using SDK)
const cursorMemory = createCortexMemory({
  convexUrl: process.env.CONVEX_URL!,
  memorySpaceId: "shared-workspace", // Same space
  userId: currentUser.id,
  agentId: "cursor-agent",
  hiveMode: { participantId: "cursor" },
});

// Claude Desktop (if using SDK)
const claudeMemory = createCortexMemory({
  convexUrl: process.env.CONVEX_URL!,
  memorySpaceId: "shared-workspace", // Same space
  userId: currentUser.id,
  agentId: "claude-agent",
  hiveMode: { participantId: "claude-desktop" },
});

// They can all see each other's conversations
const sharedMemories = await webAppMemory.search('user preferences');
// Returns memories from all three participants

2. Multi-Agent Systems

// Agent A
const agentA = createCortexMemory({
  memorySpaceId: "team-workspace",
  userId: "user-123",
  agentId: "analyzer-agent",
  hiveMode: { participantId: "analyzer" },
});

// Agent B
const agentB = createCortexMemory({
  memorySpaceId: "team-workspace",
  userId: "user-123",
  agentId: "writer-agent",
  hiveMode: { participantId: "writer" },
});

// Both see all memories
const allMemories = await agentA.search('project status');
// Returns memories from both analyzer and writer

// But can filter by participant
const analyzerMemories = await agentA.search('project status', {
  participantId: 'analyzer' // Filter to only analyzer's memories
});

Filtering by Participant

You can filter memories and facts by participantId to see data from specific tools or agents:

// Filter memories by participant
const calendarMemories = await cortexMemory.search('meetings', {
  participantId: 'calendar-tool'
});

// Filter facts by participant
const emailFacts = await cortex.facts.list({
  memorySpaceId: 'shared-workspace',
  participantId: 'email-tool'
});

Benefits

Unified Context

Share memory seamlessly across your entire AI ecosystem

Multi-Agent Coordination

Enable multiple agents to work together with shared context

Real-time Updates

Changes from one participant are immediately available to all

Participant Tracking

Filter and attribute memories by which tool or agent stored them

Limitations

When using Hive Mode, be aware of the following:

• participantId is optional: You can use hive mode without specifying a participantId, though it's recommended for tracking purposes.
• No data isolation: All participants can see all data in the shared memory space. participantId is for attribution and filtering, not access control.
• Attribution, not security: The participantId field tracks who stored data but does not provide any security or access control. All participants have full access to all memories in the shared space.

Next Steps