Email security@reddb.io privately. Do not file a public GitHub issue for security-impacting bugs.

You can also use GitHub Security Advisories for private coordinated disclosure.

We acknowledge reports within 48 business hours, confirm or reject within 5 business days, and target a fix or mitigation within 90 days for non-critical issues. Critical issues (data loss, auth bypass, RCE) are prioritized and patched as fast as we can ship and verify a fix.

In scope:

• The RedDB engine (/src, /wal, /drivers/*).
• Official Docker images under ghcr.io/reddb-io/reddb.
• Helm chart under /charts/reddb.
• The managed Cloud at *.reddb.io.

Out of scope:

• Third-party AI provider APIs (report to the provider).
• User misconfiguration (e.g. exposing port 5000 to the public internet without auth).
• Social engineering, physical attacks, denial-of-service against managed Cloud.

We follow coordinated disclosure: we ask reporters to keep findings private until a fix is available or 90 days elapse, whichever comes first. We will publicly credit reporters in the CHANGELOG and on https://reddb.io/security unless they prefer to stay anonymous.

• Run RedDB behind a reverse proxy with TLS terminated. The bundled HTTP server supports TLS but most operators prefer to terminate at the proxy.
• Set REDDB_AUTH=true and REDDB_REQUIRE_AUTH=true in production (the official images set both). Anonymous access is convenient for local development only; REDDB_NO_AUTH=true opts out explicitly.
• Set RED_ADMIN_TOKEN for the operator surface (/admin/*, /metrics). Without it those routes require an admin-role bearer from the user auth store.
• Use the vault module for AI provider keys instead of plaintext config.
• Enable WAL fsync (default). Only disable on disposable nodes where replay-from-source is acceptable.
• Restrict the admin API (/v1/admin) to a private network or behind SSO.
• Set REDDB_BOOTSTRAP_TOKEN (or REDDB_BOOTSTRAP_TOKEN_FILE) on any store whose bootstrap endpoint is reachable over the network. Bootstrap is first-caller-wins; with the token set, POST /auth/bootstrap and the gRPC AuthBootstrap call must present it in x-reddb-bootstrap-token.
• Terminate TLS on every listener that is not bound to loopback. The server warns at boot for each plaintext listener reachable beyond the host.

RedDB has no telemetry, usage reporting or update check: the server never contacts a RedDB-operated endpoint. The only outbound connections it opens are the ones an operator configures — AI providers (red.config.ai.*), the audit-log and operator-event webhook sinks, replication peers, remote backup backends, and the red ui bundle download — plus image references that a vision-enrichment policy reads from row data, which are subject to the same egress rules as AI providers (https only off loopback, no private or link-local addresses unless REDDB_AI_ALLOW_PRIVATE_PROVIDERS=1; local files only with REDDB_AI_VISION_ALLOW_LOCAL_FILES=1).

The engine is source-available under the Business Source License 1.1. To audit:

git clone https://github.com/reddb-io/reddb.git
cd reddb
cargo audit              # dependency CVEs
cargo test --workspace   # baseline correctness

Pointers for a focused review:

• Storage and durability: crates/reddb-file, crates/reddb-server/src/storage.
• Auth and tokens: crates/reddb-server/src/auth (vault in auth/vault.rs).
• Network surface: crates/reddb-server/src/server, crates/reddb-server/src/grpc, crates/reddb-server/src/wire.
• Dependency policy: Cargo.lock is committed; cargo audit runs in CI.

For a longer write-up — encryption, backups, isolation, compliance roadmap — see https://reddb.io/security.