The LLM switchboard that bills users, runs code, and knows 22 APIs

What it does Chats is both a React-based chat interface and an API gateway for large language models. It normalizes access to over 22 providers—including OpenAI, DeepSeek, Moonshot, and DashScope—behind standard Chat Completions and Gemini-style endpoints. The project bundles user management, model pricing controls, request tracing, and multimodal support into a single deployable unit.

The interesting bit Most open-source chat frontends stop at the UI; this one adds an accountant and a bouncer. It tracks per-user model usage down to the request source—web chat, API, or summary generation—supports Keycloak SSO and SMS login, and can spin up a Docker sandbox to let models execute code, browse pages, or edit Excel files. That turns it from a toy into something an internal IT team might actually approve.

Key highlights

• Normalizes 22+ providers behind OpenAI-compatible and Gemini protocol endpoints, including reasoning-model interleaved thinking
• Enterprise access controls: Keycloak SSO, SMS verification, user quotas, rate limiting, and audit logging
• Built-in code interpreter via Docker sandbox with browser automation and spreadsheet tools
• Request Trace observability for debugging inbound and outbound API calls
• Flexible backends: SQLite, SQL Server, or PostgreSQL; file storage spans local disk, S3, Aliyun OSS, and Azure Blob
• Recent releases added automatic conversation titling, a model pricing dashboard, and lazy-loaded Markdown to cut initial page weight roughly in half

Caveats

• Documentation and release notes are primarily in Chinese; English support exists but appears secondary
• The Docker quickstart requires mounting the host Docker socket, which grants the container significant privileges—worth noting for security-conscious deployments

Verdict Teams running mixed-provider LLM fleets in corporate or shared environments should look here; solo hackers who just want a quick ChatGPT clone will find the permission-management overhead excessive.