Sicura is a local security workbench that helps developers investigate exposed credentials and PostgreSQL row-level security misconfigurations.

A database policy can look correct while still allowing one user to access another user’s data. Sicura makes that behavior visible by comparing intended permissions with recorded observations in disposable local PostgreSQL replicas.

Developers can import a GitHub repository or local project files, review findings, and define an Expected Access Model. Sicura then runs controlled verification scenarios using synthetic data and low-privilege identities. For supported repairs, developers review and approve a suggested migration, apply it to a local replica, and repeat the same scenarios and regression checks before exporting the migration.

Gemma 4 E4B runs locally through Ollama to explain findings, propose expectations, and suggest constrained remediation intents. Its advisory analysis stays separate from verified database evidence. Application code controls executable actions, and AI proposals require human review.

Credential detection is static: secrets are redacted, and discovered credentials are never used. Verification requires no production database credentials.

Sicura brings detection, explanation, verification, and reviewed remediation into one workflow. Its current scope covers credential exposure and supported PostgreSQL RLS scenarios; automatic repair currently supports owner-based SELECT policies.