01Comprehensive AI Governance and Auditability: Centralized logging and tracking of all AI requests, scope decisions, tool calls, model outputs, and costs in a single Postgres table for simplified auditing.
02Prevent Secret Exfiltration: Denies tool calls attempting to pass secrets through arguments, with credentials injected into child processes without entering the LLM context.
03Self-hosted, Air-gap Capable, Single Artifact: Operates as one Rust binary and PostgreSQL instance, without external dependencies like Redis or Kubernetes, suitable for air-gapped environments.
04Policy-as-Code & Role-Based Access Control (RBAC): Implements destructive operation blocklists, department scoping, and six-tier RBAC through PreToolUse hooks, with auditable deny reasons.
05Provider-Agnostic LLM Inference Gateway: Routes /v1/messages requests from clients (e.g., Claude for Work) to various upstream providers (Anthropic, OpenAI, Gemini, Moonshot, Qwen, MiniMax) based on model patterns.
068 GitHub stars
