Context Memory Fabric (CMF) is an experimental, self-hosted developer preview designed for single-user personal deployments.

If you discover a security vulnerability or sensitive data leak risk within Context Memory Fabric:

• Do NOT create a public GitHub issue.
• Please report vulnerabilities directly and privately by emailing: info@toddmargolis.net
• Include detailed reproduction steps, the affected versions or commits, and any relevant sanitized traces.
• Reports will be reviewed promptly.

1. Single-User, Isolated Deployment:

   • Every CMF instance is completely self-contained. Running CMF locally or via Docker Compose operates against your own isolated FalkorDB graph, local files, and configured API keys.
   • Your data is never transmitted to any central CMF server, multi-tenant database, or external project telemetry.

2. Model Provider Data Flow:

   • When using Google Gemini for episodic extraction and embeddings, text sent to memory tools (remember, get_context, recall_mem) is transmitted to Google's API under your own API key terms.
   • When configured with CMF_LLM_PROVIDER=local, inference remains entirely on your local machine or local network endpoint.

3. MCP Network Exposure:

   • The streamable-http/sse network transport exposes an HTTP endpoint (default port 8000).
   • If binding beyond 127.0.0.1 (such as via a public reverse proxy or tunnel), you MUST configure CMF_MCP_AUTH_TOKEN or OAuth 2.1 authentication (CMF_MCP_ISSUER_URL and CMF_MCP_OAUTH_PASSWORD). Running unauthenticated on public interfaces allows any caller full read/write access to your memory graph and documents.