grid-memory 1.2.2

Grid Memory — shared persistent memory for multi-agent teams

Grid Memory — Python SDK

Shared persistent memory for multi-agent teams. Works with AutoGen, CrewAI, and LangGraph.

pip install grid-memory

Quick Start

from grid_memory import Grid

grid = Grid("http://localhost:8080", default_agent_id="my-agent")

# Write facts, decisions, and handoffs
grid.fact("PostgreSQL pool max: 25 connections", tags=["database"])
grid.decide("Use Express over Fastify", tags=["architecture"],
            rationale="Better middleware ecosystem maturity")
grid.handoff(from_agent="researcher", to_agent="builder",
             content="API spec ready in docs/api-v2.md", status="ready")

# Query by tags, agent, type
entries = grid.query(tags=["database"])
for e in entries["entries"]:
    print(f"[{e['type']}] {e['agent_id']}: {e['content'][:100]}")

# Inject context into agent prompts
context = grid.inject("building the API layer")
print(context)
# ─── SHARED MEMORY GRID ───
# [fact] 18:05 — agent:architect — database
# PostgreSQL pool max: 25 connections
# ─── END GRID ───

# Admin
info = grid.info()
grid.prune()
grid.forget("grid_20260529_abc123")

Per-Agent Identity

All convenience methods accept an agent_id parameter:

grid.fact("Important note", agent_id="researcher-1")   # override per-call
grid.decide("Architecture decision", agent_id="lead-architect")

Or set a default for the client:

grid = Grid(default_agent_id="worker-bee")
grid.fact("My fact")  # uses "worker-bee"

Framework Plugins

AutoGen

from grid_memory import AutoGenGridPlugin
import autogen

# Create the plugin
grid_plugin = AutoGenGridPlugin(agent_id="researcher")

# Create your agent normally
agent = autogen.AssistantAgent(
    name="researcher",
    llm_config=llm_config,
)

# Wrap it — auto-injects Grid context before messages, logs exchanges
agent = grid_plugin.wrap(agent)

CrewAI

from grid_memory import CrewAITool
from crewai import Agent

grid_tool = CrewAITool(agent_id="researcher")

agent = Agent(
    name="Researcher",
    tools=[
        grid_tool.query_tool(),    # query the Grid
        grid_tool.write_tool(),    # write to the Grid
        grid_tool.context_tool(),  # inject context
    ],
)

LangGraph

from grid_memory import langgraph_grid_node
from langgraph.graph import StateGraph

# Create a Grid injection node
inject_context = langgrid_grid_node(agent_id="my-agent")

graph = StateGraph(AgentState)
graph.add_node("grid_inject", inject_context)
graph.add_edge("grid_inject", "next_node")

Development

# Install dev dependencies
pip install -e ".[dev]"

# Run tests (requires Grid server at localhost:8080)
pytest tests/

Framework Comparison

Feature	Grid Memory	LangMem	Mem0	Zep
Multi-agent shared store	✅ Built for this	❌ Per-agent only	❌ Per-user only	❌ Per-user only
Tagged + TTL'd entries	✅ Built-in	❌	⚠️ Manual	✅ Time-based
Relevance scoring	✅ Tag > type > recency	❌	⚠️	✅
Context injection block	✅ Markdown, LLM-ready	❌	❌	❌
Append-only audit	✅ Immutable entries	❌	❌	❌
Zero dependencies	✅ Pure stdlib	❌	❌	❌
Single Docker container	✅ 130MB	❌	❌	Requires DB
API	REST + Python + Node SDK	Python SDK	Python SDK	REST + SDK

Running the Grid Server

# Docker (recommended)
docker compose up -d

# Or directly with Node (requires Node >= 18)
cd ../../
node server.js

# Verify
curl http://localhost:8080/health