Niki-1337/proxy-ai

Proxy AI is an open-source gateway that intercepts prompts sent to external LLM providers, scans and sanitizes sensitive data, and rehydrates model responses. It’s intended for developer teams and security/ops groups who want to prevent secrets, PII, internal infra, or risky context from leaking to third‑party models while preserving useful code shape and auditability. Key features: OpenAI/Anthropic-compatible proxy endpoints and a local HTTPS MITM; secret, PII, infra, source-code and prompt-injection scanning; code-shape preserving sanitization with signed __GWV1_*__ placeholders and response rehydration (handles common LLM mutations); tree-sitter aware replacements; mutation scoring and warnings; basic policy engine (allow, sanitize, block, shadow, warn, route_to_local_model, disable_rehydration); PostgreSQL audit store with safe previews and hash chain; community admin console and macOS helper scripts. Community core excludes several enterprise modules (OIDC/SAML, KMS, RBAC, SIEM integrations).

このRepoのdescription / GitHub topics / READMEのAI要約に出現したAI関連キーワードをカテゴリ別に表示。各バッジは該当カテゴリの詳細ランキングへリンクします。