shim

Teams sending prompts, documents and data to AI models like OpenAI, Anthropic or Gemini risk exposing personal data such as national IDs, IBANs, emails and phone numbers to third-party providers. This creates compliance exposure under regimes like KVKK, GDPR, DORA and the EU AI Act, and leaves companies without visibility into who called which model, what it cost, or what data was sent. Tools like LiteLLM route requests across providers but were not built to protect the data inside those requests. Without redaction or an audit trail, organizations have no verifiable evidence of what left their systems or how AI spend breaks down by key, team or project.

shim sits between an app or developer and any AI provider as a drop-in gateway, requiring only a base URL swap to redact personal data before it reaches the model and restore it in the response, so the model only ever processes placeholder tokens while the mapping stays on shim's gateway. It detects identifiers like national IDs, IBANs, emails and secrets in Turkish and English, converts PDF, Word and Excel files into clean Markdown to cut token usage, and offers model allowlisting, regional pinning and provider fallback. It also produces an exportable audit log of every call, policy in force and redaction made, and supports flagging requests like credit decisions for human review. The product ships as a free open-source CLI for individual developers, a hosted AI Gateway for platform teams, a self-hosted Enterprise version with audit and budget controls, and a browser-based checker for AI chat tools still in early access via waitlist.