# Isartor

A single-binary Rust gateway that caches and routes LLM requests from coding tools such as Copilot, Claude, and Cursor across multiple providers, aiming to deflect most redundant traffic before it reaches the cloud.

Isartor sits between coding tools and language model providers as a local gateway, applying exact and semantic caching, local small-model routing, and context compression to reduce how much traffic actually reaches paid cloud models. It connects to Copilot, Claude, Cursor, OpenClaw, Codex, Gemini, and generic OpenAI-compatible clients, and supports multi-provider fallback chains, named model aliases, and quota-aware key rotation with cooldowns after rate limits.

It ships as a single Rust binary that can run fully air-gapped with no hidden telemetry, and provider credentials are stored either as static keys or through an encrypted local credential store, with device-flow login for providers such as Copilot, Gemini, and Kiro. A guided setup command walks through configuring providers, running a demo, and connecting a coding tool, and the gateway accepts several different client request formats side by side.

Website: https://github.com/isartor-ai/Isartor