Skip to content

RFC: Cryptographic audit receipts for MindsDB agent queries #12363

@tomjwxf

Description

@tomjwxf

Problem

MindsDB enables AI agents to query live data across databases, APIs, and ML models. These queries can affect production systems, but there is no cryptographic audit trail proving which queries were executed, what policy governed them, and that the records haven't been altered.

Proposal

Add optional Ed25519 receipt signing to MindsDB's query execution layer. Each AI-driven query would produce a signed receipt with: query hash, data source, policy evaluation result, and a cryptographic signature. Receipts enable tamper-evident audit trails for regulatory compliance.

Reference

protect-mcp implements this for MCP tool calls (MIT, npm v0.5.3). Receipt format: IETF Internet-Draft. Active integrations: Mission Control, Cedar for Agents, Microsoft AGT, LlamaIndex.

Happy to discuss and contribute.

Metadata

Metadata

Assignees

No one assigned

    Labels

    No labels
    No labels

    Type

    No type

    Projects

    No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions