MemoryOps is currently in alpha. Security fixes are applied to the latest commit on main only.

Please do not open a public GitHub issue for security vulnerabilities.

Report vulnerabilities privately via GitHub Security Advisories.

Include:

• Description of the vulnerability and potential impact
• Steps to reproduce or proof-of-concept
• Any suggested mitigations

You will receive an acknowledgement within 72 hours. We aim to triage and release a fix within 14 days for critical issues.

For deployment-specific hardening, use the production checklist in docs/security-production.md.

• Workspace API keys are hashed with Argon2 before storage. The plaintext key is returned once on creation — it cannot be recovered.
• Bootstrap keys (POST /v1/workspaces) have elevated privileges. Rotate them after initial setup using POST /v1/workspaces/{id}/keys.
• Keys are passed via X-API-Key header — always use HTTPS in production.

• All webhook sources (GitHub, Slack, Jira, Linear) are validated with HMAC-SHA256 signatures.
• The dev-placeholder secret in .env.example is intentionally weak. Always set real secrets in production.
• Timing-safe comparison is used for all HMAC checks.

• No secrets are stored in config.toml. All credentials are read from environment variables at runtime.
• Never commit .env to source control. .env is in .gitignore.
• In production, use a secrets manager (AWS Secrets Manager, HashiCorp Vault, etc.) to inject environment variables rather than a .env file.

• The bootstrap endpoint (POST /v1/workspaces) should be restricted to trusted networks and disabled with WORKSPACE_CREATION_ENABLED=false after initial setup.
• The MCP server port (default 3003) should not be exposed publicly — it is intended for local AI client connections only.
• Qdrant and Redis should not be exposed to the public internet. Bind them to localhost or a private network interface.

• Use a dedicated Postgres role with least-privilege permissions for the DATABASE_URL connection.
• Enable Postgres SSL (sslmode=require) in production.
• All user-supplied values are parameterized via sqlx — no string interpolation in SQL.

• LLM and embedding API keys (OPENAI_API_KEY, ANTHROPIC_API_KEY, etc.) are read from environment variables and never logged or persisted to the database.
• Audit your LLM provider's data retention policies if you are processing sensitive memory content.