Cortex Memory

Open-source AI memory for agents 🧠 ACID + Vector hybrid. Built on @convex. Apache 2.0. Give your AI infinite context.

🔗

[Cortex Memory @CortexMemoryDev]

🎄 Happy Holidays from Cortex! 🎁

Building AI agents that never forget?

We're here to help. Star the repo, join discussions, share your use case.

Together, we're making AI memory accessible to everyone.

#HappyHolidays #Community

[Cortex Memory @CortexMemoryDev]

🔒 Security by design

✅ CodeQL scanning
✅ Dependency scanning (Trivy)
✅ Secret detection (Gitleaks)
✅ OWASP Top 10 (Semgrep)
✅ OpenSSF Scorecard
✅ Automated PR checks

Enterprise-grade security. Always.

#Security #DevSecOps

[Cortex Memory @CortexMemoryDev]

🎯 Three retrieval tiers explained

Tier 1: Facts (8 tokens each) - Most efficient
Tier 2: Summaries (75 tokens) - More detail
Tier 3: Raw context (full messages) - Complete fidelity

Adaptive retrieval = right context at right cost.

#Architecture #Optimization

[Cortex Memory @CortexMemoryDev]

💼 Why companies choose Cortex

1. Single database (no polyglot complexity)
2. ACID audit trails
3. Automatic versioning
4. Built-in governance
5. Open source core
6. Self-hostable

Enterprise-ready from Day 1.

#Enterprise #AI

[Cortex Memory @CortexMemoryDev]

🧠 User Profiles + GDPR Cascade

```typescript
await cortex.users.delete(
'user-123',
{ cascade: true }
);
```

Deletes across ALL layers:
• Conversations
• Memories
• Facts
• Graph
• Immutable
• Mutable

One call. Complete compliance.

#GDPR #Privacy

[Cortex Memory @CortexMemoryDev]

⚡ Search performance at scale

1M+ memories: <50ms retrieval
Vector dimensions: Any (768-3072+)
Concurrent users: Unlimited
Token cost: Constant

Scale doesn't slow you down. That's the Convex advantage.

#Performance #Scalability

[Cortex Memory @CortexMemoryDev]

🔥 Apache 2.0 = Enterprise friendly

✅ Free for commercial use
✅ Explicit patent protection
✅ No vendor lock-in
✅ Modify freely
✅ Same as Convex

Not just open source. SAFE open source.

#OpenSource #License

[Cortex Memory @CortexMemoryDev]

📈 Real-world use case: Customer Support

500 tickets over 5 years = 500K tokens accumulated

With Cortex: Search returns 5 most relevant facts from 5 YEARS in 40 tokens.

Perfect recall. Minimal cost. Happy customers.

#CustomerSupport #AI

[Cortex Memory @CortexMemoryDev]

🔧 MCP Server = Memory everywhere

Install once → Works in:
• Cursor IDE
• Claude Desktop
• Custom tools
• Any MCP client

Your AI memory follows you across ALL applications.

#MCP #CrossPlatform

[Cortex Memory @CortexMemoryDev]

🎨 Framework agnostic = true freedom

Cortex works with:
• LangChain ✅
• Vercel AI SDK ✅
• LlamaIndex ✅
• OpenAI ✅
• Anthropic ✅
• Your custom code ✅

One memory system. Any AI framework.

#Integration #Flexibility

[Cortex Memory @CortexMemoryDev]

💡 Pro tip for chatbots

Don't accumulate conversation history!

❌ Send all 500 past messages every time
✅ Retrieve top 5 relevant memories

Constant tokens. Infinite scale. Better results.

This is the "Infinite Context" pattern.

#ChatbotDev #AI

[Cortex Memory @CortexMemoryDev]

🔗 Context Chains for multi-agent systems

Supervisor can see subordinate context.
Specialist reads manager's workflow.
Clean hierarchies. No information silos.

Read-only. Limited. Audited. Secure.

Perfect for complex agent orchestration.

#MultiAgent #Architecture

[Cortex Memory @CortexMemoryDev]

⚡ Fact Extraction = 60-90% storage savings

One conversation exchange:
→ Raw: ~300 tokens
→ Summarized: ~75 tokens
→ Facts: ~8 tokens per fact

1000 exchanges = Retrieve 80 tokens instead of 25,000

Same context. Massive savings.

#Optimization #AI

[Cortex Memory @CortexMemoryDev]

Did you know? 🤔

Cortex automatically versions your memories!

Every update preserves history (10 versions default). Never lose data.

Perfect for:
• Audit trails
• Debugging
• Compliance
• Temporal queries

#DataManagement #AI

[Cortex Memory @CortexMemoryDev]

🔒 Your data. Your infrastructure.

Cortex NEVER sends data outside your Convex deployment.

• Deploy to Convex Cloud ☁️
• Run localhost 💻
• Self-host anywhere 🏢

100% control. Zero vendor lock-in.

#Privacy #OpenSource

[Cortex Memory @CortexMemoryDev]

🎯 When to use Graph DB with Cortex

Perfect for:
✅ Deep context chains (5+ levels)
✅ Knowledge graphs with relationships
✅ Multi-hop reasoning
✅ Provenance tracking

Not needed for:
❌ Simple chatbots
❌ Basic RAG

Choose the right tool for your use case.

#GraphDB #AI

[Cortex Memory @CortexMemoryDev]

💰 Cost comparison for 1000 exchanges:

Traditional (accumulation): $1.50/request
Summarization: $0.30/request
Cortex (facts only): $0.0024/request
Cortex (hybrid): $0.012/request

Up to 99% savings. Use smaller models. Same results.

#CostOptimization #AI

[Cortex Memory @CortexMemoryDev]

🔍 Multi-strategy search FTW

Cortex automatically tries:
1️⃣ Semantic vector search (best results)
2️⃣ Keyword text search (always works)
3️⃣ Recent memories (contextual fallback)

You always get results. No empty responses.

#Search #AI

[Cortex Memory @CortexMemoryDev]

🏢 One-click GDPR compliance

```typescript
const policy = await cortex
.governance
.getTemplate('GDPR');

await cortex
.governance
.setPolicy(policy);
```

That's it. Enterprise compliance solved.

Also: HIPAA, SOC2, FINRA templates built-in.

#Compliance #Enterprise

[Cortex Memory @CortexMemoryDev]

⚡ Real talk about embeddings

You don't need a 3072-dim model for everything!

768 dims: High-volume, cost-sensitive ✅
1536 dims: General purpose ✅
3072 dims: When accuracy is CRITICAL ✅

Choose based on your use case, not "bigger = better"

#MachineLearning #VectorSearch