Anthropic - AI Rooms Workflow Addon

Overview

Claude-powered AI agent addon for Rooms AI, providing advanced conversational AI, file analysis, and web-enhanced responses.

Addon Type: anthropic ( 'agent' type addon )

Features

• Chat Completion: Advanced conversational AI using Claude models with tool integration
• File Analysis: Upload and analyze documents, images, and other files with Claude
• Web Search: Enhanced responses with real-time web search capabilities and citations
• Tool Registry: Full support for external tools and workflow integration
• Token Management: Comprehensive token tracking and usage monitoring

Add to Rooms AI using poetry

Using the script

poetry add git+https://github.com/synvex-ai/anthropic-rooms-pkg.git

In the web interface, follow online guide for adding an addon. You can still use JSON in web interface.

Configuration

Addon Configuration

Add this addon to your AI Rooms workflow configuration:

{
  "addons": [
    {
      "id": "claude-assistant",
      "type": "anthropic",
      "name": "Claude AI Assistant",
      "enabled": true,
      "config": {
        "model": "claude-3-5-sonnet-20241022",
        "max_tokens": 4096,
        "temperature": 0.7
      },
      "secrets": {
        "anthropic_api_key": "ANTHROPIC_API_KEY"
      }
    }
  ]
}

Configuration Fields

BaseAddonConfig Fields

All addons inherit these base configuration fields:

Field	Type	Required	Default	Description
id	string	Yes	-	Unique identifier for the addon instance
type	string	Yes	-	Type of the addon ("template")
name	string	Yes	-	Display name of the addon
description	string	Yes	-	Description of the addon
enabled	boolean	No	true	Whether the addon is enabled

CustomAddonConfig Fields (anthropic-specific)

This anthropic addon adds these specific configuration fields:

Field	Type	Required	Default	Description
model	string	No	"claude-3-5-sonnet-20241022"	Claude model to use
max_tokens	integer	No	4096	Maximum tokens for responses
temperature	float	No	0.7	Temperature for text generation (0.0-1.0)

Required Secrets

Secret Key	Environment Variable	Description
anthropic_api_key	ANTHROPIC_API_KEY	Anthropic API key for Claude access

Environment Variables

Create a .env file in your workflow directory:

# .env file
ANTHROPIC_API_KEY=your_anthropic_api_key_here

Available Actions

chat_completion

Advanced conversational AI using Claude models with full tool integration support.

Parameters:

• message (string, required): User message to send to Claude
• messages (array, optional): Full conversation history with role/content pairs
• max_tokens (integer, optional): Override config default for max tokens
• temperature (float, optional): Override config default for temperature
• system (string, optional): System prompt to guide Claude's behavior

Output Structure:

• response (string): Claude's response text
• model (string): Model used for the response
• usage (object): Token usage information (input_tokens, output_tokens, total_tokens)
• stop_reason (string): Why Claude stopped generating (end_turn, max_tokens, etc.)

Workflow Usage:

{
  "id": "ai-conversation",
  "name": "Chat with Claude",
  "action": "claude-assistant::chat_completion",
  "parameters": {
    "message": "{{payload.user_question}}",
    "system": "You are a helpful assistant specialized in technical documentation."
  }
}

file_analysis

Upload and analyze documents, images, and other files with Claude's multimodal capabilities.

Parameters:

• message (string, required): Question or instruction about the file
• file_upload (object, optional): File to upload and analyze
  • file_path (string, required): Path to file to upload
  • filename (string, optional): Custom filename
  • purpose (string, default: "analysis"): Purpose of file upload
• file_id (string, optional): ID of already uploaded file
• max_tokens (integer, optional): Override config default
• temperature (float, optional): Override config default

Output Structure:

• response (string): Claude's analysis response
• file_info (object): Information about the analyzed file
• model (string): Model used
• usage (object): Token usage information

Workflow Usage:

{
  "id": "document-analysis",
  "name": "Analyze Document", 
  "action": "claude-assistant::file_analysis",
  "parameters": {
    "message": "Summarize the key points in this document",
    "file_upload": {
      "file_path": "{{payload.document_path}}",
      "purpose": "analysis"
    }
  }
}

web_search

Enhanced responses with real-time web search capabilities and source citations.

Parameters:

• query (string, required): Search query or question
• max_tokens (integer, optional): Override config default
• temperature (float, optional): Override config default
• system (string, optional): System prompt

Output Structure:

• response (string): Claude's response enhanced with web information
• citations (array): List of sources Claude referenced
• search_performed (boolean): Whether web search was actually performed
• model (string): Model used
• usage (object): Token usage information

Workflow Usage:

{
  "id": "web-enhanced-response",
  "name": "Web-Enhanced Answer",
  "action": "claude-assistant::web_search", 
  "parameters": {
    "query": "{{payload.user_question}}",
    "system": "Provide comprehensive answers with citations from recent sources."
  }
}

Tools Support

This agent addon supports tools in the chat_completion action only.

Tool Integration

The addon includes a comprehensive tool registry that allows external tools to be registered and used during Claude conversations. Tools are automatically passed to Claude when available.

Tool-Supported Actions:

• chat_completion: Full tool integration - Claude can call external tools during conversation

Using Tools with Claude

When tools are registered via useStorage or useContext, they become available to Claude during chat_completion:

{
  "id": "claude-with-tools",
  "name": "Claude with Database Tools",
  "action": "claude-assistant::chat_completion", 
  "useStorage": {
    "addonId": "my-mongo-db",
    "action": [
      {"name": "describe", "description": "Describe the database"},
      {"name": "insert", "description": "Insert a new document into the database"}
    ]
  },
  "parameters": {
    "message": "Please analyze our user database and create a summary report",
    "system": "You are a data analyst with access to database tools. Use them as needed."
  }
}

For detailed tool usage patterns, refer to the AI Features documentation.

Testing & Lint

Like all Rooms AI deployments, addons should be roughly tested.

A basic PyTest is setup with a cicd to require 90% coverage in tests. Else it will not deploy the new release.

We also have ruff set up in cicd.

Running the Tests

poetry run pytest tests/ --cov=src/template_rooms_pkg --cov-report=term-missing

Running the linter

poetry run ruff check . --fix

Pull Requests & versioning

Like for all deployments, we use semantic versioning in cicd to automatize the versions.

For this, use the apprioriate commit message syntax for semantic release in github.

Developers / Mainteners

• Adrien EPPLING : adrienesofts@gmail.com