Agent System

Backflow includes an AI agent system for workflows, embeddings, and LLM interactions.

Enable Agent System

{
  "agent": {
    "enabled": true,
    "embedding": {
      "provider": "openai",
      "model": "text-embedding-3-small",
      "apiKey": "{{env.OPENAI_API_KEY}}"
    },
    "vectorStore": {
      "provider": "qdrant",
      "url": "{{env.QDRANT_URL}}",
      "apiKey": "{{env.QDRANT_API_KEY}}",
      "collection": "documents"
    },
    "llm": {
      "provider": "anthropic",
      "model": "claude-sonnet-4-20250514",
      "apiKey": "{{env.ANTHROPIC_API_KEY}}"
    }
  }
}

LLM Providers

OpenAI

{
  "llm": {
    "provider": "openai",
    "model": "gpt-4o",
    "apiKey": "{{env.OPENAI_API_KEY}}"
  }
}

Anthropic

{
  "llm": {
    "provider": "anthropic",
    "model": "claude-sonnet-4-20250514",
    "apiKey": "{{env.ANTHROPIC_API_KEY}}"
  }
}

LMStudio (Local)

{
  "llm": {
    "provider": "lmstudio",
    "baseUrl": "http://localhost:1234/v1",
    "model": "local-model"
  }
}

Google Gemini

{
  "llm": {
    "provider": "google",
    "model": "gemini-2.0-flash",
    "apiKey": "{{env.GOOGLE_API_KEY}}"
  }
}

Embeddings

Generate Embeddings

POST /agent/embeddings
Content-Type: application/json

{
  "texts": ["Hello world", "Another text"]
}

Store Document

POST /agent/embeddings/document
Content-Type: application/json

{
  "document": "Full document text...",
  "contentType": "article",
  "metadata": {
    "source": "web",
    "url": "https://example.com"
  }
}

Search Documents

POST /agent/embeddings/search
Content-Type: application/json

{
  "query": "search query",
  "limit": 10,
  "threshold": 0.7
}

LLM Streaming

Stream Chat

POST /llm/stream
Content-Type: application/json

{
  "messages": [
    {"role": "user", "content": "Hello!"}
  ],
  "stream": true
}

With System Prompt

POST /llm/stream
Content-Type: application/json

{
  "messages": [
    {"role": "system", "content": "You are a helpful assistant."},
    {"role": "user", "content": "What is TypeScript?"}
  ]
}

Agent Endpoints

Method	Endpoint	Description
GET	/agent/health	Agent system health
POST	/agent/embeddings	Generate embeddings
POST	/agent/embeddings/document	Store document
POST	/agent/embeddings/search	Search documents
POST	/agent/embeddings/batch	Batch operations
POST	/llm/stream	LLM streaming chat
GET	/llm/models	List available models

MCP Tools

Connect to MCP (Model Context Protocol) servers:

{
  "agent": {
    "mcpServers": [
      {
        "name": "filesystem",
        "command": "npx",
        "args": ["-y", "@modelcontextprotocol/server-filesystem", "/path"]
      }
    ]
  }
}

List Tools

GET /agent/mcp/tools

Execute Tool

POST /agent/mcp/execute
Content-Type: application/json

{
  "server": "filesystem",
  "tool": "read_file",
  "args": {
    "path": "/path/to/file.txt"
  }
}

Vector Store Providers

Qdrant

{
  "vectorStore": {
    "provider": "qdrant",
    "url": "https://your-cluster.qdrant.io:6333",
    "apiKey": "{{env.QDRANT_API_KEY}}",
    "collection": "documents"
  }
}

Qdrant (Local)

# Start Qdrant
npm run qdrant

{
  "vectorStore": {
    "provider": "qdrant",
    "url": "http://localhost:6333",
    "collection": "documents"
  }
}