Elysia-Api
Elysia-API is a self-hosted multi-upstream AI model gateway that centralizes authentication, routing, load balancing, format conversion (OpenAI/Claude/Gemini), streaming, usage analytics, and WebUI management.
Ficha actualizada 23 sept 2026, 17:11 · upstream 23 sept 2026, 16:40 · release v1.4.0
Qué es Elysia-Api
Descripción editorial construida a partir del repositorio, su documentación y sus señales de mantenimiento.
Elysia-API is a self-hosted multi-upstream AI model gateway. It centralizes request authentication, model-group routing, load balancing, OpenAI / Claude / Gemini format conversion, streaming forwarding, usage statistics collection, and WebUI management in a single Go backend application.
The architecture uses Maheshvara as the core protocol layer for handling requests across different upstream models while maintaining consistent internal representation. The WebUI is embedded directly into the binary using Go's embed package, serving at /ui/ by default with React + Vite frontend. Runtime configuration is managed via a bootstrap config.json file that can be hot-reloaded.
Data persistence is handled through SQLite for model sources, usage logs, and system metrics. The gateway supports per-key scheduling of multiple API keys within a single source instance, each with independent permission discovery and auto-discovery of available models per key.
- Centralized AI model orchestration where organizations need to route requests across multiple LLM providers (OpenAI, Claude, Gemini) through a unified gateway
- Production-grade API proxy that handles authentication, rate limiting, and format conversion for heterogeneous client applications
- Model-as-a-Service (MaaS) platform with built-in usage analytics, token tracking, and per-model capability discovery
- Secure model routing infrastructure where different API keys are scheduled to different upstream models with automatic permission-based model selection
- AI agent orchestration layer that converts various LLM response formats into a standardized internal format for downstream processing pipelines
- Self-hosted inference gateway with WebUI management, allowing developers to manage and monitor multiple AI endpoints from a single dashboard
- Maheshvara core protocol provides consistent request/response representation across OpenAI, Claude, and Gemini formats while preserving semantic meaning
- Built-in WebUI embedded in the binary with React + Vite frontend, eliminating the need for separate deployment infrastructure
- Per-key scheduling with per-key model permission auto-discovery ensures no unauthorized access to restricted models
- Streaming responses converted via stateful Maheshvara decoders support SSE output for all built-in and custom protocols
- Usage statistics track cache hits, reasoning tokens, multimodal token usage, tool calls, and retry events at the model level
- Model-level management with individual enable/disable/delete operations and preserving merge of manually added models
- The Maheshvara protocol is proprietary to Elysia and not part of any open standard; upstream providers must support it for full compatibility
- SQLite persistence means the database grows unbounded over time, requiring periodic cleanup or migration strategies
- Per-key model permission auto-discovery requires each key to request a fresh list of available models, which can be slow under high concurrency
- The default port 8765 may conflict with existing services on the same host without proper configuration management
- No built-in circuit breaker or exponential backoff for upstream failures; manual intervention is required for degraded performance
Perfil de infraestructura
Estimación inicial derivada del repositorio y de los servicios detectados.
Despliegue detectado
Qué señales reales de instalación hemos encontrado.
Install dependencies with `npm install`. Build the WebUI, synchronize embedded assets, and cross-compile standalone binaries using `npm run build` to generate platform-specific artifacts. For production deployment, use Docker Compose with a minimal configuration that mounts the SQLite database volume (`elysia-data:/data`) and sets `ELYSIA_API_HOST=0.0.0.0`. Enable hot-reload by setting `ELYSIA_API_RELOAD=true` in environment variables. Rotate the panel access token after each login session to prevent unauthorized access. Monitor usage statistics via the WebUI dashboard for model performance and cost estimation.
Stack y dependencias
Componentes inferidos a partir de manifests, configuración y documentación.
Puede sustituir a
Relaciones detectadas por el análisis del repositorio y su categoría.
Experiencia real de la comunidad
Ayuda a contrastar la estimación automática con instalaciones reales.
Todavía no hay datos comunitarios para este proyecto.