DeepEye orchestrates LLM-assisted workflows, uploaded data, database connections, generated artifacts, and Docker-backed execution runtimes. Treat the current repository as an active development preview unless you have reviewed and hardened it for your own deployment environment.

For a component-level trust boundary overview, see docs/security_model.md.

Security fixes are currently targeted at the default branch.

Please do not disclose exploitable details in public issues, pull requests, or discussions.

Preferred reporting path:

1. Use GitHub private vulnerability reporting or GitHub Security Advisories for this repository if enabled.
2. If private reporting is unavailable, open a minimal public issue asking for a secure contact path without including exploit details.

Include the affected component, impact, reproduction conditions, and any suggested mitigation when using a private channel.

Please review these areas carefully before exposing DeepEye outside a trusted local environment:

• Docker socket and runtime-control access
• sandbox lifecycle, filesystem mounts, and cleanup
• generated report, dashboard, and video code paths
• LLM tool execution and prompt-to-action boundaries
• uploaded files and database connection strings
• authentication, refresh tokens, cookies, CORS, and gateway configuration
• MinIO, Postgres, Redis, and local volume persistence

Before production-like use:

• replace all example secrets and development credentials
• use least-privilege service accounts and network boundaries
• isolate Docker/runtime control from public API surfaces
• set CPU, memory, storage, and execution time limits
• restrict outbound network access for generated-code workloads where possible
• enable audit logging and container cleanup policies
• review dependency and container image vulnerability scans

Recommended local checks:

make check
make security-scan