ThinkWatch-Lite

Overview of ThinkWatch-Lite

What it is

ThinkWatch Lite is a local gateway for AI clients like Claude Code and Codex, available on macOS, Windows and Linux. It sits between the client and the upstream AI provider, intercepting and optionally modifying requests and responses without altering the client's configuration after the initial setup.

Key features

• Connect clients once, then switch upstreams and models solely in the gateway without touching client configuration.
• Secrets & privacy replaces API keys, private keys, JWTs, passwords, Chinese ID numbers and bank card numbers with placeholders before requests leave the machine.
• Dangerous commands cuts off tool calls that download and run code, exfiltrate environment variables or credential files, read private keys, or install startup items or scheduled jobs.
• Upstreams serving the same model are cross-checked; anomalies such as mismatched model names or unusual reported input sizes are flagged with evidence and sample sizes on hover.
• Every request is recorded with cost and route; finished requests can be replayed against another upstream and compared side by side; the full history, including request and answer text, is searchable.
• Routing rules based on model, tools, images, extended thinking, and other criteria; on failure the next upstream takes over while each session sticks to one upstream to preserve prompt cache hits.
• Supports API keys, Amazon Bedrock, ChatGPT and Z.ai accounts, relays (e.g., OpenRouter), and local models, with conversion between Anthropic, OpenAI and Gemini API formats.

Use cases

• Operate multiple AI clients without reconfiguring each one when changing AI providers or model endpoints.
• Keep sensitive API keys and private data away from relay services by swapping them before the request leaves the local machine.
• Block dangerous or undesired tool calls inserted by a relay before the client can execute them.
• Compare responses from different upstreams for the same model to detect inconsistencies or anomalies.
• Audit and trace every request, including its cost, route, and the content of both request and answer.

Getting started

macOS (Apple silicon)bashbrew install --cask thinkwatchproject/tap/thinkwatch-liteOr download the ThinkWatch-Lite--darwin-arm64.dmg.#### Windows x64Download the installer ThinkWatch-Lite--windows-x64-setup.exe or the portable ThinkWatch-Lite--windows-x64-portable.zip.#### Linux (x86_64 or aarch64)bashcurl -fsSL https://github.com/ThinkWatchProject/ThinkWatch-Lite/releases/latest/download/install.sh | shOr download the ThinkWatch-Lite--linux-.AppImage.#### Build from sourcebashpnpm installbash src-tauri/scripts/fetch-core.shpnpm tauri dev

License and pricing

The project is licensed under MIT. It is open source; no pricing is stated in the README or metadata.

Based on the project's public README and repository metadata.