AgentMem

The memory layer for AI agents. Postgres-native, TypeScript- and Python-first, framework-agnostic. Drop it into any agent stack (LangChain, LangGraph, CrewAI, Mastra, Claude SDK, custom) in three lines.

import { MemoryStore } from '@agentmem/sdk'
 
const mem = new MemoryStore({ apiKey: process.env.AGENTMEM_API_KEY! })
 
await mem.write({ content: 'Alice prefers email over phone', agentId: 'support-bot' })
const hits = await mem.search({ query: 'how should I reach Alice?', agentId: 'support-bot' })

What you get

• Write + search with embeddings, BM25 keyword search, and entity-graph traversal — all merged with Reciprocal Rank Fusion. Optional Gemini re-ranking for high-precision queries.
• Multi-agent coordination. Scopes (private / team / global) + roles (planner / executor / observer) + access grants give you fine-grained visibility control out of the box.
• CRDT conflict resolution. Vector clocks per agent, configurable policies (planner_wins, timestamp_wins, human_in_loop), and supersession lineage so you can audit who replaced what.
• TTL + retention. Memories expire automatically. Retention policies prune old data on a schedule.
• Webhooks. Subscribe to memory.written, memory.expired, memory.conflict_detected, etc. — HMAC-signed payloads.
• BYOK encryption for compliance-sensitive deployments.

Install

# TypeScript
npm install @agentmem/sdk
 
# Python
pip install agentmem-py

Where to next

• Getting started — install, first write, first search, in 3 minutes.
• Core concepts — scopes, roles, workflows, vector clocks. Read this before you build anything non-trivial.
• Features — deep-dives into each shipped feature.
• API reference — every endpoint with curl examples.
• SDK reference — every method on the TS and Python clients.