Transparent, simple, fast, reliable, elastic —
multi-backend proxy for Claude Code
Hot-switch between multiple LLM backends without leaving Claude Code.
When launched via hi cc, the proxy automatically intercepts and routes all Claude Code API requests —
completely transparent, no files modified, restored on exit.
Single binary, cross-platform, zero dependencies.
🔒 Privacy-first by default — All Anthropic and Claude telemetry requests are automatically disabled. No usage data ever leaves your machine. Zero privacy leakage.
Prerequisite: install Claude Code first, then install hi below.
⌨️ CLI tool — launch proxy + Claude Code in one command. Auto-detect API keys from config or env vars. Proxy automatically intercepts all Claude Code API requests — no manual setup.
A transparent proxy that gives you the freedom to choose the best model for every task.
Instantly switch between Claude, DeepSeek, MiniMax, GLM, Kimi (K2.6 / K2.7-code), Qwen, and MIMO — no restart needed. Auto-generated slash commands for every model.
No files modified on disk. hi intercepts API calls transparently and restores original configuration on exit. Your Claude Code setup stays pristine.
All Anthropic and Claude telemetry requests are blocked by default when using hi cc. No usage data, no crash reports, no analytics — nothing leaves your machine.
Single static binary — download and run. Cross-platform support for Linux, macOS, and Windows. No runtime, no package manager, no hassle.
Track token usage and cost across all backends with persistent history. Know exactly what each model costs you, session after session.
Route to faster/cheaper models for simple tasks, switch to premium models for complex reasoning. Optimize cost and latency per prompt.
Auto-detect settings with hi init-config, or skip config entirely and use environment variables. Your choice.
Three steps from zero to multi-backend Claude Code.
One command via curl or irm. Single binary, no dependencies.
Run hi init-config to auto-detect API keys. Or grab the pre-built config template covering Claude and all Chinese open-source models — just fill in your keys.
Run hi — proxy starts and intercepts all API requests. Claude Code launches automatically. Telemetry blocked.
Type /claude, /deepseek, /kimi — hot-switch backends mid-session.
Pre-built binaries for Linux, macOS, and Windows. Same command for install and upgrade.
curl -fsSL https://raw.githubusercontent.com/mars-base/hi/main/install.sh | sh
⚠️ Run the following commands in PowerShell (not CMD)
irm https://raw.githubusercontent.com/mars-base/hi/main/install-windows.ps1 | iex
Or use CMD:
curl -fsSL https://raw.githubusercontent.com/mars-base/hi/main/install-windows.cmd -o install.cmd && install.cmd && del install.cmd
After install, run hi init-config to auto-detect your API keys.
Config files: ~/.hi/config.yaml (Linux/macOS) ·
%USERPROFILE%\.hi\config.yaml (Windows)
Or skip config — use env vars:
ANTHROPIC_API_KEY=sk-xxx DEEPSEEK_API_KEY=sk-xxx hi
Multiple LLM providers, one proxy. Hot-switch between them with a single slash command.