What is Backflow?

Backflow is the backend for AI-powered applications. Define AI workflows in JSON. Get production APIs, MCP tools, and SDKs—automatically.

The Problem

Building AI backends takes months. Every project requires the same infrastructure:

• Multi-provider LLM integration
• Workflow orchestration
• Streaming responses
• MCP tool servers
• Database operations
• Authentication and rate limiting

You spend weeks building infrastructure instead of shipping features.

The Solution

Backflow gives you multi-provider LLM, workflow orchestration, and production patterns—in minutes. Define your AI workflows in a simple JSON config. Backflow handles:

• Multi-Provider LLM - Claude, GPT, Gemini, local models—one unified API
• Workflow Orchestration - DAG execution with dynamic planning and auto-recovery
• MCP Native - Expose APIs as MCP tools automatically for Claude, Cursor, and AI assistants
• Real-time Streaming - SSE and WebSocket for live AI responses
• Built-in RAG - Document parsing, chunking, embeddings, and vector search
• Production APIs - Auto-generated REST endpoints with OpenAPI docs

Key Features

Multi-Provider LLM

Switch between Claude, GPT, Gemini, and local models with an env var—no code changes.

{
  "workflows": [{
    "name": "support-bot",
    "steps": [{
      "type": "llm",
      "provider": "anthropic",
      "model": "claude-sonnet-4-20250514",
      "prompt": "You are a helpful support agent."
    }]
  }]
}

MCP Native

Expose your APIs as MCP tools automatically. Claude, Cursor, and AI assistants connect directly to your backend.

Workflow Orchestration

DAG execution with dynamic planning and auto-recovery. LLM workflows that handle failures gracefully.

Built-in RAG

Document parsing, chunking, embeddings, vector search. PDF, DOCX, images with OCR—all built-in.

Production Ready

• Multi-tenancy with tenant isolation
• RBAC with permissions
• Rate limiting per endpoint
• Caching (Redis, SQLite, Memory)
• 50+ integrations (JIRA, GitHub, Stripe, Slack, and more)

When to Use Backflow

• AI-powered applications and agents
• LLM workflow backends
• MCP tool servers
• Multi-tenant SaaS
• Rapid prototyping and MVPs