Skip to main content

AI Chat Interface

OpenRegister's AI chat interface provides conversational AI capabilities with RAG (Retrieval Augmented Generation), function calling, and comprehensive agent management.

Overview​

The AI chat interface enables users to:

  • Interact with AI agents using natural language
  • Search across data using semantic search and RAG
  • Execute actions via function calling (tools)
  • Control context by selecting views and tools
  • Configure search with flexible RAG settings

Agent Selector​

The agent selector displays available AI agents in a compact card-based grid layout.

Card Layout​

Each agent card shows:

  • Agent name and description
  • Start Conversation button (positioned on the right)
  • Views: Available data views the agent can access
  • Tools: Available function calling tools
  • Model: LLM provider and model name

Key Features​

Compact Design:

  • Responsive grid layout
  • Side-by-side views and tools display
  • Expandable lists (shows first 3, expand to see all)
  • Optimized for various screen sizes

Visual Hierarchy:

  • Clear agent identification
  • Immediate call-to-action visibility
  • Capability overview at a glance
  • Model information displayed

Chat Settings​

Access chat settings via the settings button next to the send button.

View and Tool Selection​

Views Section:

  • Checkboxes for all available views from the agent
  • Controls which data the AI can access
  • All views checked by default
  • Toggle items on/off during conversation

Tools Section:

  • Checkboxes for all available tools from the agent
  • Controls which actions the AI can execute
  • All tools checked by default
  • Dynamic tool management

RAG Configuration​

Include Toggles:

  • Include Objects: Search in structured object data
  • Include Files: Search in document files

Source Count Controls:

  • Object Sources: 1-20 (default: 5)
  • File Sources: 1-20 (default: 5)

Performance Guidance:

  • Fewer sources (1-3): Faster responses, more focused
  • More sources (10-20): Comprehensive context, slower responses
  • Recommended: 5 sources for balanced speed and accuracy

Function Calling​

Function calling allows AI agents to execute actions and retrieve data.

Available Tools​

  • Application Tool: List, create, and manage applications
  • Agent Tool: List, create, and manage agents
  • Register Tool: List, create, and manage registers
  • Schema Tool: List, create, and manage schemas
  • Object Tool: Search, create, and manage objects

Provider Support​

ProviderFunction Calling Support
OpenAI✅ Fully supported
Ollama✅ Fully supported
Fireworks AI❌ Not yet implemented

Note: To use function calling, switch to OpenAI or Ollama in LLM Configuration settings.

RAG (Retrieval Augmented Generation)​

RAG enables AI agents to search and retrieve relevant context from your data.

How RAG Works​

  1. User sends message to AI agent
  2. System searches vectors using semantic similarity
  3. Retrieves relevant sources (objects and files)
  4. Builds context from retrieved sources
  5. AI generates response using context

Search Modes​

Semantic Search:

  • Uses vector embeddings
  • Finds semantically similar content
  • Works across languages

Hybrid Search:

  • Combines keyword and semantic search
  • Uses Reciprocal Rank Fusion (RRF)
  • Best accuracy and coverage

Keyword Search:

  • Traditional text search
  • Fast and reliable
  • Good for exact matches

Configuration​

Configure RAG in chat settings:

  • Select which views to search
  • Choose object/file inclusion
  • Set source count limits
  • Control search behavior

Use Cases​

Speed-Optimized Queries​

  • Object sources: 2
  • File sources: 2
  • Minimal views/tools enabled
  • Result: Fast responses (~1-2s)

Comprehensive Research​

  • Object sources: 15
  • File sources: 15
  • All relevant views enabled
  • Result: Thorough analysis (~5-10s)
  • Include Objects: ✓
  • Include Files: ✗
  • Only data-related tools enabled
  • Result: Structured data queries
  • Include Objects: ✗
  • Include Files: ✓
  • Document tools only
  • Result: Document/policy questions

API Reference​

Send Message​

POST /apps/openregister/api/chat/send

Request:

{
    "message": "Your question here",
    "conversation": "conversation-uuid",
    "views": ["view-uuid-1", "view-uuid-2"],
    "tools": ["tool-uuid-1", "tool-uuid-2"],
    "includeObjects": true,
    "includeFiles": true,
    "numSourcesFiles": 5,
    "numSourcesObjects": 5
}

Response:

{
    "message": {
        "id": 123,
        "content": "AI response",
        "sources": [...]
    },
    "conversation": "conversation-uuid",
    "title": "Conversation Title"
}

Tools & RAG Implementation​

Tools Fixed​

All tools now use consistent OpenAI function calling format:

public function getFunctions(): array
{
    return [
        [
            'name' => 'list_applications',
            'description' => 'List all applications...',
            'parameters' => [
                'type' => 'object',
                'properties' => [
                    'limit' => [
                        'type' => 'integer',
                        'description' => 'Maximum results',
                    ],
                ],
                'required' => [],
            ],
        ],
    ];
}

Function Calling Integration​

Function calling is integrated into the chat service:

  • Tools are loaded from agent configuration
  • Converted to function definitions
  • Passed to LLM via setTools() method
  • Function calls are executed and results returned to LLM

Provider Support​

ProviderFunction Calling Support
OpenAI✅ Fully supported
Ollama✅ Fully supported
Fireworks AI❌ Not yet implemented

Note: To use function calling, switch to OpenAI or Ollama in LLM Configuration settings.

RAG Status​

RAG is fully operational with:

  • ✅ Semantic search finds objects correctly
  • ✅ Source names extracted from metadata
  • ✅ Vector metadata includes uuid, register, schema, file_id, file_path
  • ✅ Cosine similarity calculation
  • ✅ Hybrid search (SOLR + vectors + RRF)
  • ✅ View-based filtering
  • ✅ Object and file vectorization

Configuration​

Using Function Calling:

  1. Switch to OpenAI or Ollama in LLM Configuration
  2. Enable tools for agent
  3. Test with natural language commands

Using RAG:

  1. Vectorize your data (objects and files)
  2. Configure agents with appropriate views
  3. Test queries to verify RAG is finding correct information
  4. Monitor vector stats to track coverage