We release security updates for the following versions of Cortex:

Note: Cortex is currently in alpha. Once we reach 1.0, we will support:

• Latest major version (security + features)
• Previous major version (security only, for 6 months)

If you discover a security vulnerability, please DO NOT open a public GitHub issue. Public disclosure before a fix is available puts all users at risk.

Report security vulnerabilities through one of these channels:

1. Go to our Security Advisories page
2. Click "Report a vulnerability"
3. Fill out the form with detailed information

Send an email to: [email protected]

Include:

• Type of vulnerability
• Full description of the issue
• Steps to reproduce
• Potential impact
• Suggested fix (if any)
• Your contact information

Use our PGP key for highly sensitive reports:

-----BEGIN PGP PUBLIC KEY BLOCK-----
[PGP Key will be added here]
-----END PGP PUBLIC KEY BLOCK-----

Key fingerprint: [Fingerprint will be added]

1. Acknowledgment: Within 48 hours
2. Initial Assessment: Within 5 business days
3. Updates: Every 5-7 days until resolution
4. Fix Timeline:
   • Critical: 7-14 days
   • High: 14-30 days
   • Medium: 30-60 days
   • Low: Next release cycle

We will:

• Acknowledge your report within 48 hours
• Keep you informed of our progress
• Credit you in the security advisory (unless you prefer anonymity)
• Work with you to understand and validate the issue
• Provide a CVE identifier if applicable

// ❌ DON'T: Expose Convex credentials in client-side code
const cortex = new Cortex({
  convexUrl: "https://your-deployment.convex.cloud",
  // Never put API keys in client-side code!
});

// ✅ DO: Use environment variables on the server
const cortex = new Cortex({
  convexUrl: process.env.CONVEX_URL,
});

// ❌ DON'T: Trust user input directly
await cortex.memory.remember(userInput); // Could contain malicious data

// ✅ DO: Validate and sanitize
import { z } from "zod";

const RememberInputSchema = z.object({
  memorySpaceId: z.string().regex(/^agent-[\w-]+$/),
  conversationId: z.string(),
  userMessage: z.string().max(10000),
  agentResponse: z.string().max(10000),
  userId: z.string(),
  userName: z.string(),
  importance: z.number().min(0).max(100), // 0-100 scale
});

const validated = RememberInputSchema.parse(userInput);
await cortex.memory.remember(validated);

// ❌ DON'T: Allow unrestricted access
app.post("/api/memory", async (req, res) => {
  await cortex.memory.remember(req.body); // No auth check!
});

// ✅ DO: Verify user permissions
app.post("/api/memory", requireAuth, async (req, res) => {
  // Verify user owns this agent
  if (!(await userOwnsAgent(req.user.id, req.body.agentId))) {
    return res.status(403).json({ error: "Forbidden" });
  }

  // Store with Layer 3 remember()
  await cortex.memory.remember({
    memorySpaceId: req.body.agentId,
    conversationId: req.body.conversationId,
    userMessage: req.body.userMessage,
    agentResponse: req.body.agentResponse,
    userId: req.user.id,
    userName: req.user.name,
  });
});

// Embeddings can be sensitive - treat them as private data
const embedding = await openai.embeddings.create({
  model: 'text-embedding-3-large',
  input: userContent, // May contain PII
});

// Store securely (Layer 2 for system-generated content)
await cortex.vector.store(agentId, {
  content: userContent,
  contentType: 'raw',
  embedding: embedding.data[0].embedding,
  source: { type: 'system', timestamp: new Date() },
  metadata: {
    importance: 85,  // 0-100 scale (70-89 = high)
    // Don't leak user info in metadata
  },
});

// Or use Layer 3 for conversations (handles ACID + Vector)
await cortex.memory.remember({
  agentId, conversationId, userMessage, agentResponse, userId, userName,
  generateEmbedding: async (content) => await openai.embeddings.create({...})
});

// Protect against abuse with rate limiting
import rateLimit from "express-rate-limit";

const memoryLimiter = rateLimit({
  windowMs: 15 * 60 * 1000, // 15 minutes
  max: 100, // Limit each IP to 100 requests per window
  message: "Too many requests, please try again later",
});

app.use("/api/memory", memoryLimiter);

// Log security-relevant actions
const result = await cortex.memory.remember({
  agentId,
  conversationId,
  userMessage,
  agentResponse,
  userId,
  userName,
});

// Log to audit trail
await auditLog.record({
  action: "MEMORY_STORED",
  userId: req.user.id,
  agentId,
  conversationId: result.conversation.conversationId,
  acidMessageIds: result.conversation.messageIds, // ACID layer
  vectorMemoryIds: result.memories.map((m) => m.id), // Vector layer
  timestamp: new Date(),
  ip: req.ip,
});

• Never commit secrets or credentials
• Use TypeScript strict mode
• Validate all inputs
• Use parameterized queries
• Sanitize outputs
• Follow principle of least privilege

# Regularly check for vulnerabilities
npm audit

# Fix automatically when possible
npm audit fix

# Review manual fixes
npm audit fix --force

All code changes must:

• Pass automated security checks
• Be reviewed by at least one maintainer
• Include tests for security-relevant features
• Document security implications

Issue: Agent memories must be properly isolated.

Mitigation:

• Cortex enforces agent-level isolation at the database level
• All queries filter by agentId
• Cross-agent queries are explicitly blocked

Issue: Embeddings can potentially leak information about the original text.

Mitigation:

• Treat embeddings as sensitive data
• Don't expose raw embeddings in APIs
• Use proper access control
• Consider encryption at rest for sensitive deployments

Issue: Search timing could potentially reveal information.

Mitigation:

• Search operations have consistent timing
• Results are limited to authorized agents
• Implement rate limiting

Issue: Cortex depends on Convex's security model.

Mitigation:

• Follow Convex Security Best Practices
• Use Convex authentication
• Implement row-level security in Convex functions
• Regular security audits

1. Day 0: Vulnerability reported
2. Day 2: Initial response sent
3. Day 7: Severity assessment complete
4. Day 14-90: Fix developed and tested (depending on severity)
5. Day 90: Public disclosure (or earlier if fix is deployed)

We prefer coordinated disclosure:

• We'll work with you on a disclosure timeline
• We'll credit you in the advisory
• We may request embargo until users can update
• We'll publish a security advisory once fixed

After a fix is deployed, we will:

1. Publish a GitHub Security Advisory
2. Release a patched version
3. Update the CHANGELOG with security notes
4. Notify users via Discord and Twitter
5. Request a CVE if severity warrants

Get notified of security updates:

1. GitHub: Watch our repository and enable security alerts
2. Discord: Join #security-announcements channel
3. Email: Subscribe to security mailing list at [email protected]
4. RSS: Subscribe to our security feed

We use CVSS 3.1 for severity ratings:

• Critical (9.0-10.0): Immediate action required
• High (7.0-8.9): Update within 7 days
• Medium (4.0-6.9): Update within 30 days
• Low (0.1-3.9): Update when convenient

Currently, we do not have a formal bug bounty program. However:

• We greatly appreciate security reports
• We will credit you in our security advisories
• We may send Cortex swag as a thank you
• We're considering a formal program for v1.0+

Cortex is designed to support compliance with:

• GDPR: Data portability, deletion, and privacy by design
• CCPA: User data rights and deletion
• SOC 2: When deployed with appropriate controls
• HIPAA: With proper Convex configuration (contact us for guidance)

Note: Compliance ultimately depends on how you deploy and configure Cortex.

• Convex Security Documentation
• OWASP Top 10
• CWE Top 25
• Security Headers

• Security Team: [email protected]
• General Inquiries: [email protected]
• Emergency: [email protected] (include "URGENT" in subject)

We thank the following security researchers for responsibly disclosing vulnerabilities:

No vulnerabilities reported yet.

Last Updated: 2025-10-23
Version: 1.0

Thank you for helping keep Cortex and our community safe! 🔒