Engrammic
Guides

Multi-Agent Memory

Share knowledge across multiple agents

One agent learns, all agents remember.

When multiple agents connect to the same Engrammic endpoint, they share a single knowledge graph. Knowledge stored by one agent is instantly available to every other agent, with no syncing and no hand-offs.

Example: A code agent discovers the project uses pnpm workspaces. Later, a review agent retrieves that same fact without ever being told, because it came from the shared graph.

Setup

Connect agents to the same endpoint

Each agent's MCP config points to the same URL:

{
  "mcpServers": {
    "engrammic": {
      "url": "https://beta.engrammic.ai/mcp/"
    }
  }
}

Both agents now share the same knowledge graph.

Team Workflows

Common patterns for multi-agent setups:

  • Code + Review: Code agent stores discoveries, review agent recalls them for context
  • Research + Implementation: Research agent stores findings, implementation agent applies them
  • Specialist handoff: Each specialist contributes to shared knowledge others can query

Avoiding conflicts: When multiple agents might update the same knowledge, recall first to check current state, then store with supersedes pointing to the existing node.

On this page