faz launches safety layer for AI agents accessing databases, enforcing a 5-stage inspection pipeline before queries execute.

faz sits between AI agents (Claude, Cursor, or MCP clients) and databases, routing every query through RBAC, AST analysis, injection detection, and guardrails before execution.

Agents connect via MCP or REST API and see four tools: list_databases, describe_table, query, and federated_query (which joins results across multiple databases). Each query returns either results or a detailed block reason from the safety stage that rejected it.

Supports 14 databases across relational (PostgreSQL, MySQL, Oracle), document (MongoDB, CouchDB), search (Elasticsearch, OpenSearch), vector (Weaviate, Qdrant, Milvus, Pinecone), graph (Neo4j), wide-column (Cassandra), and cloud (DynamoDB) categories.

Install via pip (pip install faz-core), configure permissions in faz.yaml with per-table read/write/append rules, and auto-configure Claude Desktop or Cursor with faz mcp install.