Group Memory

Shared memory across agents and teammates. When one agent learns something important, it becomes available to everyone in the group.

Overview

Group Memory extends Agent Memory to work across multiple agents and human teammates. This enables:

• Knowledge sharing — Discoveries propagate across the team
• Collective learning — The group gets smarter as individuals learn
• Onboarding acceleration — New agents inherit existing knowledge
• Reduced redundancy — No need to re-learn what others already know

How It Works

// Agent A discovers a pattern

rlm_remember({

  content: "The payment API requires idempotency keys",

  type: "fact",

  scope: "team",  // Share with the team

  category: "backend"

})

// Agent B (different session) can recall it

rlm_recall({

  query: "What do I need for payment API calls?",

  scope: "team"

})

// Returns: "The payment API requires idempotency keys"

Memory Scopes

Memories can exist at different scopes:

Creating Groups

Groups can be created through the dashboard or programmatically. Each group has a unique ID and can contain multiple agents and human members.

Via Dashboard

1. Go to Dashboard → Agents → Groups
2. Click Create Group
3. Enter a name and optional description
4. Invite members (agents or humans)

Via MCP Tool

Use the rlm_swarm_create MCP tool to create a group programmatically:

rlm_swarm_create({

  name: "team-backend",

  description: "Backend team agents",

  max_agents: 10

})

MCP Tools for Groups

Storing Group Memories

Use the scope parameter to control visibility:

rlm_remember({

  content: "Use snake_case for database column names",

  type: "decision",

  scope: "team",

  category: "coding-standards"

})

Recalling Group Memories

When recalling, specify the group to include shared memories:

rlm_recall({

  query: "What are our database naming conventions?",

  scope: "team",

  category: "coding-standards"

})

Listing Team Memories

Use rlm_memories with scope filtering to browse team memories:

rlm_memories({

  scope: "team",

  category: "coding-standards",

  limit: 20

})

Memory Inheritance

When querying memories, the system checks multiple scopes in order of specificity:

Results are merged and ranked by relevance, with agent memories given a slight boost to preserve personalization.

Use Cases

Team Coding Standards

Store coding conventions as group rules so all agents follow the same patterns:

rlm_remember({

  content: "All API responses must include a 'success' boolean field",

  type: "decision",

  scope: "team",

  category: "api-standards"

})

Shared Discoveries

When one agent discovers something useful, share it immediately:

// Agent finds a bug workaround

rlm_remember({

  content: "The stripe SDK has a bug with null metadata - always pass empty object",

  type: "learning",

  category: "bugs",

  scope: "team"

})

Project Context

Store project-wide facts that all agents should know:

rlm_remember({

  content: "This project uses Next.js 14 with App Router",

  type: "fact",

  scope: "project"  // Available to all project agents

})

Best Practices

• Use team scope for shared knowledge — Don't over-share personal preferences
• Use categories for organization — Group related memories with meaningful category names
• Keep scopes focused — Use "project" for project-specific, "team" for cross-project knowledge
• Review group memories periodically — Prune outdated information
• Combine with Snipara Context — Use memories for dynamic knowledge, Context for static docs

Plan Limits

Next Steps