Your AI agents keep forgetting.
Let's fix that.
Production-grade semantic memory for AI agents. Works with Python, TypeScript, and any tool via MCP.
Wait, what does this actually do?
The Problem
Talk to ChatGPT today, and tomorrow it has no idea who you are. AI apps start fresh every conversation — like a goldfish with amnesia.
The Solution
memable gives your AI a "notebook" it can actually write to and read from. It remembers users, preferences, past decisions — whatever matters.
Why It Matters
Your AI gets smarter over time. It provides personalized experiences, remembers context, and builds relationships — instead of starting from zero.
Think of it like this:
Without memable
"Hi! I'm your AI assistant. What's your name?"
(asks this every. single. time.)
With memable
"Hey Sarah! Last week you asked about React performance. Want to continue that?"
(actually remembers you)
One memory, every tool
Claude Desktop, Cursor, your custom app — they all share the same memory via MCP. Learn something in one place, remember it everywhere.
Built for real products
From support bots to personal assistants — anywhere AI needs to remember.
Customer Support
Remember past tickets, preferences, and conversation history. No more "Can you explain your issue again?"
Personal Assistants
Learn routines, preferences, and relationships over time. Your AI that actually knows you.
Sales & CRM
Track deal context, past conversations, and buying signals. AI that remembers every touchpoint.
Education & Tutoring
Adapt to learning styles, track progress, remember where students struggle.
Gaming NPCs
Characters that remember player choices, build relationships, and hold grudges.
E-commerce
Personalized recommendations that improve over time. Remember sizes, styles, past purchases.
Works everywhere
Python, TypeScript, or any tool via MCP
from memable import build_store, MemoryCreate
with build_store("postgresql://...") as store:
# Remember something
store.add(namespace, MemoryCreate(
text="User prefers dark mode",
durability="core", # Never forget
))
# Recall it later
memories = store.search(namespace, "user preferences")Add to your config file:
Built for production
Not another demo. Actually works.
Version Chains
When facts change, we create an audit trail. Debug why your agent thinks what it thinks.
Durability Tiers
Core facts (forever), situational context (temporary), episodic memories (decay over time).
Multi-Backend
Postgres for production, SQLite for dev, DuckDB for analytics. No vendor lock-in.
Memory Types
Facts, rules, decisions, preferences, observations. Structured semantics for smart retrieval.
Edge-Ready
TypeScript client works with Neon serverless on Cloudflare Workers. Sub-100ms latency.
Auto-Extraction
Pass in conversations, get structured memories out. LLM-powered extraction finds facts, preferences, and decisions automatically.
Memory for any tool
The included MCP server lets Claude Desktop, Cursor, and other tools remember things across sessions. No code changes needed.
Don't want to manage your own database?
Interested in a hosted option? Let me know