FootyVision is a research/portfolio project that runs locally: Postgres in Docker, a local LLM server, and an API bound to localhost. It has no authentication layer and is not hardened for public deployment. Do not expose the API to the internet as-is.

Please open a private security advisory rather than a public issue. I aim to respond within a week.

• No raw SQL from user input. Natural-language search is constrained to a validated Pydantic PlayerQuery; the SQL is built by trusted code with bound parameters.
• No secrets in the repo. .env is gitignored; .env.example holds local defaults only.
• No outbound data. Chat and embeddings go to an OpenAI-compatible endpoint you run yourself; no player data leaves the machine.
• Container images run as a non-root user and verify TLS when installing dependencies.

• The API is unauthenticated and not rate-limited, by design (local use).
• LLM output is grounded in computed statistics but is still model-generated prose — it is not a substitute for verifying the underlying numbers, which the API exposes directly.