Help & Documentation

Header Controls

Left Sidebar

• Quick Actions — Grouped shortcuts for Tools, Vector Store, Database, Documents, and Help
• MCP Standard Prompts — Pre-built prompts from the MCP server
• My Prompts — Create and save your own reusable prompts
• Chat Sessions — History of all conversations, search, create new session

Chat Area

• Type your message and press Enter to send
• Use Shift+Enter for new lines
• Click the paperclip icon to upload files (PDFs, images) for OCR processing
• Use the export button to save the conversation

Status Bar

The bottom status bar shows: MCP connection status, Circuit Breaker state, active LLM provider/model, current orchestration layer, token count, and a Clear Cache button.

Navigation Menu

Click the hamburger menu (desktop) to access: AI Agents, Models, Tools & Data, Applications, Platforms, Monitoring, Developer Tools, and App Settings.

Core Agents 5 agents

KPI Agents 6 agents

Specialist Agents 8+ agents

• Data Discovery — Explore and catalog available data sources
• Database Objects Explorer — Deep-dive into tables, views, stored procedures
• OCR Document Specialist — Extract text and data from scanned documents
• SQL Optimization — Query tuning, explain plans, performance advice
• RAG Pipeline — Retrieval-Augmented Generation configuration
• Chat Completions — General conversational AI interactions
• TDVS (Vector Store) — Teradata vector store management
• Backup & Restore — Database backup operations and recovery
• Compliance Reviewer — Regulatory compliance checks
• DB Space Explorer — Storage analysis and space management

How Agent Routing Works

When you send a query, the L1 Supervisor automatically selects the best agent based on keyword matching and semantic understanding. You do not need to specify which agent to use — just ask your question naturally.

How It Works

When agents return data that is suitable for charting, the auto-visualization pipeline kicks in automatically. It detects the data type, selects the best chart format, and renders an interactive chart in your chat.

Supported Chart Types (16+)

Fraud Network Visualization

The Risk Management agent can generate network graphs showing fraud relationships. These use ECharts force-directed layouts to visualize customer-transaction connections, highlighting high-risk nodes.

Charts Panel & Dashboard

• Toggle the Charts Panel via the bar-chart icon in the header to browse all generated charts
• Multi-chart responses automatically render as a fullscreen dashboard
• Each chart is interactive: hover for tooltips, click legend items to filter

Quick Actions Sidebar (5 Categories)

My Prompts

Create, save, and reuse your own custom prompts. Click "My Prompts" in the sidebar, then use the "+" button to create a new prompt. Saved prompts persist across sessions.

MCP Standard Prompts

The MCP server provides pre-built prompts for common operations. Access them from the sidebar under "MCP Standard Prompts".

Standalone OCR Page

Visit /ocr for the dedicated OCR processing page, also linked from the landing page.

In-Chat File Upload

Click the paperclip icon in the chat input to upload PDFs or images. After upload, a document context banner appears with quick actions:

• Summarize — Get a summary of the document
• Extract Data — Pull structured data from the document
• Find Tables — Locate and extract tabular data
• Identify — Identify document type and key information

OCR Engine Selection

Settings Tabs (17 total)

Open Settings from the gear icon in the header. Available tabs:

• LLM Provider, Teradata, Vector Store, ModelOps, Context Memory
• File Paths, Extended Thinking, Performance, Multi-Agent, RAG Config
• Monitoring, Cost Management, Security, UI Preferences
• Prompts, Platform Connections, Cache Management

Configuration Priority

1. UI Settings (highest) — Changes made in the Settings panel
2. .env file — Environment variables on the server
3. Code defaults (lowest) — Hardcoded in config.py

LLM Providers

Platform Connections Tab

Test connectivity to all infrastructure services: Redis, PostgreSQL, Prometheus, Grafana, Ollama, ChromaDB. Also includes Docker Compose controls for starting/stopping services.

Option A — Certificate Import (Recommended)

1. Download the certificate:
   Download langgraph-ca.crt (or visit http://187.77.180.86:5000/static/langgraph-ca.crt)
2. Import into Chrome:
   • Windows: Open chrome://settings/security → Manage certificates → Or press Win+R → certmgr.msc → Trusted Root Certification Authorities → Certificates → right-click → All Tasks → Import → select the .crt file → Place in "Trusted Root Certification Authorities" → Finish → confirm the security warning → Restart Chrome
   • Mac: Open Keychain Access → double-click the cert → expand "Trust" → set to "Always Trust"
   • Linux: Open chrome://settings/security → Manage certificates → Authorities tab → Import → select the file → check "Trust this certificate for identifying websites"
3. Verify: Visit https://187.77.180.86:2025/ok — you should see {"ok":true} without a security warning.
4. Open Agent Studio:
   https://smith.langchain.com/studio/?baseUrl=https://187.77.180.86:2025
   Look for the green "Connected" badge.

Option B — Chrome Flags (Quick Alternative)

1. Open chrome://flags/#unsafely-treat-insecure-origin-as-secure
2. Paste http://187.77.180.86:2024 in the text box
3. Set to Enabled → Relaunch Chrome
4. Open Studio: https://smith.langchain.com/studio/?baseUrl=http://187.77.180.86:2024

Connecting to Neo4j Browser

1. Open http://187.77.180.86:7474
2. In the connection form, change the protocol dropdown from neo4j:// to bolt://
3. Set URL to 187.77.180.86:7687
4. Username: neo4j  |  Password: aifactory
5. Click Connect

Sample Queries

How the Knowledge Graph Works

Entities are automatically extracted from every AI conversation and stored as nodes in Neo4j. Relationships between entities are created and maintained automatically. Entity resolution runs every 30 minutes to merge duplicate entities.

Slow Responses?

Enable the Fast toggle in the header to activate Token Streaming. This delivers the first tokens in ~500ms instead of waiting 8-14s for the complete response.

Wrong Agent Selected?

Use the Layer Override dropdown in the header to force a specific orchestration layer (L1/L2/L3). L1 is the multi-agent supervisor with the best routing.

Stale Cache?

Click Clear Cache in the status bar at the bottom. For a full cache reset, the server admin can call POST /api/cache/clear {"target": "semantic"}.

MCP Disconnected?

Check the MCP status indicator in the bottom status bar. If the circuit breaker has tripped (shown in red), click the reset button next to it. The circuit breaker auto-recovers after a cooling period.

Charts Not Appearing?

Toggle the Charts Panel using the bar-chart icon in the header. Charts are rendered inside the panel and inside chat messages. If a chart still does not appear, try clearing the cache and resending the query.

Need a Clean Start?

• Clear the chat: Ctrl+L (or Cmd+L on Mac)
• Clear the cache: click "Clear Cache" in the status bar
• Start a new session: click "New Session" in the sidebar