CoAI.Dev is a self-hosted platform that combines an end-user chat application with an LLM gateway and billing backend, positioned as a single deployment covering both consumer-facing chat and API distribution. The Go backend and React frontend expose an OpenAI-compatible API (/v1/chat/completions, /v1/images, /v1/models, /v1/billing) in front of multiple upstream providers, so one site can serve both a chat UI and API keys for developers.
Upstream providers are managed as channels. Channels carry priorities that decide call order, weights that distribute load among channels of equal priority, user-group scoping, automatic retry on failure, model redirection, and status management. A response cache keyed on the request-parameter hash returns prior results without billing when enabled.
The chat side includes cross-device conversation sync and sharing, file parsing (PDF, Docx, Pptx, Excel, images with OCR) via a companion blob service backed by S3, R2, or MinIO, web search through SearXNG, text-to-image via DALL-E, Midjourney, and Stable Diffusion, and PWA and Tauri desktop builds. The admin dashboard handles users, subscriptions, elastic per-request or per-token billing, gift and redemption codes, pricing, a customizable model market, announcements, and SMTP settings.
Supported providers include OpenAI and Azure OpenAI, Anthropic Claude, Google Gemini, Midjourney, iFlytek SparkDesk, Zhipu ChatGLM, Alibaba Tongyi Qwen, Tencent Hunyuan, Baichuan, Moonshot, DeepSeek, ByteDance Skylark, Groq, OpenRouter, 360 GPT, and LocalAI or Ollama through the OpenAI format. A commercial edition adds TTS and STT, a plugin marketplace, RAG knowledge bases, more payment and auth providers, channel health detection with failover, multi-tenant API key distribution, rate limiting, and chat-bot integrations.
Features
- OpenAI-compatible proxy: chat completions, image generation, model list, and billing endpoints
- Channel management: priority ordering, weighted load balancing, user grouping, retry, model redirection, and upstream hiding
- Billing modes: subscriptions plus elastic per-request, per-token, free, or anonymous billing with minimum-points checks
- Response caching: cache hits by request hash with configurable TTL and result count, not billed
- File parsing: PDF, Docx, Pptx, Excel, and images with OCR for any model, via the blob service
- Web search: SearXNG-backed search across Google, Bing, DuckDuckGo, Wikipedia, Arxiv, and others with safe-search and truncation options
- Text-to-image: DALL-E, Midjourney with U/V/R actions, and Stable Diffusion
- Conversation sync and sharing: cross-device sync, link sharing, and save-as-image
- Model market and presets: admin-defined model descriptions and tags, user presets with cloud sync
- Admin dashboard: users, subscriptions, gift and redemption codes, pricing, announcements, site branding, and SMTP
- Upstream sync: pull channel, model-market, and pricing settings from an upstream site
- Clients: responsive web UI, PWA, and Tauri desktop, with light and dark themes and multiple interface languages
