Dukememory is local-first software for storing agent memory, retrieval metadata, code context, and audit events. Treat it as part of your local development environment, not as a hardened multi-tenant hosted service.

Open a private GitHub security advisory if available, or contact the repository owner through GitHub. Please include:

• affected commit or release;
• reproduction steps;
• expected impact;
• whether local files, memory rows, embeddings, hooks, or MCP transport are involved.

Do not publish exploit details before there is time to assess and patch.

Dukememory assumes:

• a single-user local workstation by default;
• local PostgreSQL with pgvector;
• local Ollama models;
• project-scoped memory lookup;
• explicit user intent for cross-project access.

Important safety properties:

• new memory writes pass a local safety policy before insertion;
• obvious secret values are blocked before memory writes;
• possible personal data can be reported as warning-level findings;
• automatic extraction writes pending memories by default;
• MCP tools accept project_path so clients can bind reads to the active workspace root;
• temporal graph facts can be invalidated without deleting audit history.

Dukememory does not currently claim:

• multi-tenant isolation;
• hardened network exposure;
• complete secret detection;
• cryptographic protection of local database contents;
• safety guarantees for untrusted model output.

Keep the database, generated backups, exported JSON files, and Codex hook configuration under the same access controls as your source code.