deeix-ai/deeix-chat

DEEIX Chat is an open source workspace that lets a team use many AI models from many vendors in one place. The README calls it an enterprise AI workspace and lists chat, image generation, file workflows, tool use, billing, identity, and admin operations as part of the same product. The frontend is built with Next.js 16 and React 19, and the backend is a Go 1.25 service. The chat side supports multi-branch conversations, streaming replies, retries, edits, feedback, sharing, file attachments, and traces that show how each request was executed. A separate image flow handles generation and editing using an OpenAI Images compatible protocol, with its own task history kept apart from text chat. Protocols listed for upstream models include OpenAI Responses and Chat Completions, OpenAI Images, Anthropic Messages, Google Gemini Generate Content, xAI Responses, OpenRouter defaults, and custom OpenAI compatible routes. Administrators configure platform model names, real upstream channels, route bindings, priority and weight routing, pricing, and circuit breaker state through a console. The same console covers users, roles, auth providers, subscriptions, balances, usage logs, audit logs, and runtime settings. Billing supports subscription plans, top-ups, tiered pricing by token, call, or duration, free models, Stripe Checkout, EPay, and webhook validation. Identity covers local login, registration, 2FA with TOTP, recovery codes, trusted devices, and SSO through OIDC and OAuth. Files and RAG handle upload, preview, OCR, text extraction, chunking, embeddings, and semantic retrieval. The storage layer uses PostgreSQL with pgvector and Redis, with optional S3 compatible object storage. The README also mentions MCP servers for tools, structured logs, request IDs, Swagger docs, and optional OpenTelemetry. Quick start covers a local dev path with make run and pnpm dev, a lightweight Docker Compose that needs an external Postgres and Redis, and a full stack compose that brings up the app, Postgres, and Redis together. A separated deployment mode is shown for serving the frontend and backend on different public origins. The project is released under Apache 2.0.