Welcome to Bodhi App
Bodhi App is a unified AI gateway that brings together local LLM inference, cloud API proxying, and MCP tool integration under a single platform — with built-in OAuth2 authentication, role-based access control, and drop-in compatibility for the most popular AI APIs.
Run a 7B model on your laptop in the morning, route to GPT-4 or Claude in the afternoon, and let your agents call MCP tools all day — through one set of endpoints, one auth token, one UI.
Key Features
-
Multi-format API gateway Bodhi exposes the same underlying inference layer through several wire formats simultaneously:
- OpenAI —
/v1/chat/completions,/v1/models,/v1/embeddings - OpenAI Responses —
/v1/responses(async polling) for long-running and reasoning workloads - Anthropic Messages —
/anthropic/v1/messageswith first-class Anthropic OAuth (use a Claude.ai/Console account directly, no API key needed) - Gemini —
/v1beta/*for any tool that speaksx-goog-api-key - Ollama —
/api/*(deprecated, kept for legacy clients)
- OpenAI —
-
Local + remote inference, side by side Run local GGUF models via llama.cpp and proxy to remote providers (OpenAI, Anthropic, Gemini, Groq, OpenRouter, HuggingFace, or any OpenAI-compatible API). Define model aliases that combine a model file with inference parameters and use them interchangeably with API models in chat or via API.
-
MCP tool integration Connect to Model Context Protocol servers, discover their tools, and let your models invoke them mid-conversation. Three authentication methods are supported out of the box:
- Header-based — static API key or bearer token sent on every request
- OAuth2 with preregistered client — for MCP servers that publish a fixed client_id
- OAuth2 with Dynamic Client Registration (DCR) — for servers that implement RFC 7591/8414, no manual setup Includes a built-in MCP playground to test tools, an admin MCP store to publish pre-registered servers across the workspace, and an MCP proxy so external apps can route their own MCP traffic through Bodhi's auth gateway.
-
Built-in chat UI with agentic tool execution Streaming responses, markdown rendering, tool call/result inline display, parameter and system-prompt overrides, and one-click MCP tool selection per conversation.
-
Role-based auth and access control OAuth2 PKCE login with a four-tier role hierarchy (User, PowerUser, Manager, Admin). User access requests, app access requests with resource consent, and scoped API tokens for programmatic access.
-
Multi-platform Desktop apps for macOS (Intel + Apple Silicon), Windows, and Linux. Docker images for CPU (AMD64 + ARM64), CUDA, ROCm, Vulkan, MUSA, Intel, and CANN.
-
Bodhi JS SDK for developers
@bodhiapp/bodhi-js-react— register your app, request MCP access, and call OpenAI-compatible APIs from any web application.
Quick Start
- Install Bodhi App — Desktop and Docker setup with guided wizard
- Concepts → Overview — Understand the mental model before you dive in
- Features — Chat, Models, MCPs, Auth, Settings
- API Compatibility — OpenAI / Anthropic / Gemini / Ollama drop-in usage
- Developer Guide — Integrate your app with the Bodhi JS SDK
- Docker Deployment — Production deployment with GPU acceleration