This project is developed on a single main branch — there is no formal
long-term-support version matrix yet. Security fixes are applied to main
and released as soon as possible.
If you discover a security vulnerability, please do not open a public GitHub issue. Instead, report it privately by emailing:
Please include as much detail as you can:
- A description of the vulnerability and its potential impact
- Steps to reproduce, or a minimal proof-of-concept request/payload
- Any relevant logs, stack traces, or environment details
You should receive an acknowledgment within a few days. We'll work with you to understand and validate the issue, and aim to ship a fix before any public disclosure. We ask that you give us a reasonable amount of time to address the issue before disclosing it publicly.
Given this project executes LLM-generated SQL against a connected database, areas of particular interest include (but aren't limited to):
- Prompt-injection bypasses that lead to unintended query generation
- Ways to make the agent emit or execute mutating statements (
INSERT,UPDATE,DELETE,DROP,ALTER,TRUNCATE, etc.) despite the query checker and the read-only database guard - Ways to access tables/data outside the schema explicitly provided to the agent
- Any path that leaks database credentials, connection strings, or other secrets through logs or responses
Thank you for helping keep this project and its users safe.