MoonPalace is an API debugging tool from Moonshot AI designed for developers interacting with the Kimi LLM API. It acts as a local proxy, capturing, inspecting, and exporting API requests and responses to simplify troubleshooting during development and facilitate model improvement through structured feedback.

How It Works

MoonPalace operates as a local HTTP proxy. Developers redirect their application's base_url to MoonPalace (e.g., http://localhost:9988/v1). The tool intercepts all Kimi API calls, logging comprehensive details including network errors, and storing them in a local SQLite database. This enables efficient searching, inspection, and export of requests using identifiers like request_id or chatcmpl_id.

Quick Start & Requirements

Installation is via go install github.com/MoonshotAI/moonpalace@latest or downloading pre-compiled binaries. Run moonpalace start --port <PORT> (default 9988), then update your application's base_url to http://127.0.0.1:<PORT>/v1. A default API key can be set with --key.

Highlighted Details

• Cross-Platform: Supports macOS, Windows, and Linux.
• Comprehensive Logging: Captures full requests, responses, and network errors.
• Advanced Search & Export: Retrieve and export requests by request_id, chatcmpl_id, or id, with options for tagging and categorizing (good/bad cases).
• Auto-Caching: Caches /chat/completions responses based on size, TTL, and cleanup policies.
• Output Quality Monitoring: Detects content truncation and repetitive output, offering suggestions and interruption capabilities to save costs.
• Connection Stability: --force-stream option converts non-streaming requests to streaming to mitigate proxy-induced connection errors.
• Predicate Filtering: Advanced filtering of captured requests using expressions on JSON fields.

Maintenance & Community

Developed by Moonshot AI (月之暗面). No specific community channels, active maintainer details, sponsorships, or roadmap links were found in the provided README.

Licensing & Compatibility

The specific open-source license is not stated in the README. Compatibility for commercial use or closed-source linking is undetermined.

Limitations & Caveats

Future development includes using Kimi for debugging, enhanced retrieval, batch export, auto-reporting, API mocking, and a web UI. Repeat content detection is most effective with streaming output. The license remains unknown.