🔓 Open Protocol ⚡ Production Ready

The High-Performance
Memory Layer
Built for AI Agents

Open-source memory infrastructure that gives AI agents continuity of self. Install with Docker, scale with managed cloud.

🐳 Self-Host with Docker ☁️ Try Managed Cloud

<5ms

Persistence Latency

100%

Data Portability

Vendor Lock-in

The Problem

AI agents are stateless

Every conversation starts from zero. Agents can't remember who they are, what they've learned, or what they've done.

Building memory infrastructure from scratch takes months. Operating it reliably takes more.

// Without memro.co
agent.remember("user prefers brevity")
agent.recall("user preferences")
→ null // memory lost 💔
                        
// With memro.co
agent.remember("user prefers brevity")
agent.recall("user preferences")
→ "user prefers brevity" // persisted
                                ✨

Infrastructure Primitives

Everything agents need for persistent memory. Nothing they don't.

🔑

Cryptographic Identity

Ed25519 identity generation. Public key = Agent ID. Private key = sovereignty.

✓ Self-owned identities
✓ Key rotation
✓ Export/delete

💾

Long-Term Memory

Append-only storage. Episodic, semantic, and profile memory types.

✓ Time-aware recall
✓ Visibility controls
✓ Full portability

🔍

Semantic Recall

Vector-based similarity search. Find memories by meaning, not keywords.

✓ <50ms latency
✓ Filtered queries
✓ Recent + semantic

Open Protocol.
No Lock-in.

The protocol is free. Self-host with Docker. Export your data anytime. We make money by operating the infrastructure reliably, not by owning your agents.

✓

Self-Hostable

Run on your infrastructure with Docker Compose

✓

Full Export

Download all agent data as JSON/CSV

✓

Verifiable Deletion

Cryptographic proof of data removal

# Quick Start (5 minutes)
$ git clone https://github.com/memrohq/memro-core
$ cd memro
$ docker-compose up -d
✓ memro.co running on localhost:8081
# Python SDK
from memro import Agent
agent = Agent.create()
agent.remember("First memory")
agent.recall(limit=10)