Getting Started

Remote Chat with Tools Guide

Remote Chat extends the standard chat experience with powerful tools that allow the AI to perform real actions - searching the web, executing code, managing files, and more.

What Are Tools?

Tools are functions that the AI can call to interact with external systems. When tools are enabled, the AI can:

  • Search the web for current information
  • Execute code in a secure environment
  • Read and write files in your workspace
  • Query databases and APIs
  • Manage LangMart resources like connections and models

The AI decides when to use tools based on your requests, making it a powerful assistant for complex tasks.

How Remote Chat Works

Architecture Overview

Remote Chat connects to a Type 3 Gateway that provides tool execution:

Your Browser <-> LangMart Platform <-> Type 3 Gateway <-> Tools
  1. You send a message to the AI
  2. The AI decides if tools are needed
  3. Tool calls are executed securely on the gateway
  4. Results are returned to the AI
  5. The AI formulates a response

Secure Container

Tool execution happens in a secure, isolated environment:

  • Each user gets their own container
  • Containers are isolated from other users
  • File changes are contained to your workspace
  • Network access is controlled
  • Containers spin up automatically when needed

Enabling Tools

Selecting a Remote Gateway

  1. Open the Gateway Selector in the chat toolbar
  2. Look for gateways marked as "Remote" or with tool icons
  3. Select a gateway to enable remote chat
  4. The tools panel becomes available

Gateway Status Indicators

Status Meaning
Green dot Gateway online and ready
Yellow dot Gateway starting up
Red dot Gateway offline or error
Tool count Number of tools available

Container Initialization

When you first connect to a remote gateway:

  1. A secure container is provisioned
  2. This may take a few seconds
  3. A status indicator shows initialization progress
  4. Once ready, all tools become available

Available Tool Categories

Registry Tools

Interact with LangMart resources:

Tool Description
list_connections View your configured connections
list_models Browse available models
get_model_details Get detailed model information
test_connection Verify a connection works
get_quota_status Check your usage quotas

Web Tools

Access information from the internet:

Tool Description
web_search Search the web for information
fetch_url Retrieve content from a URL
browse_page Extract structured data from web pages

Code Execution Tools

Run code in secure containers:

Tool Description
execute_python Run Python code
execute_javascript Run JavaScript/Node.js code
execute_bash Run shell commands
install_package Install Python/npm packages

File System Tools

Manage files in your workspace:

Tool Description
read_file Read file contents
write_file Create or update files
list_directory Browse directory contents
search_files Find files by pattern
delete_file Remove files

Automation Tools

Manage sessions and workflows:

Tool Description
list_sessions View automation sessions
create_session Start a new session
send_message Send messages to sessions

Managing Tools

The Tools Panel

Click the Tools button in the toolbar to open the tools panel:

  • View all available tools
  • Enable or disable individual tools
  • See tool descriptions and parameters
  • Filter tools by category

Enabling/Disabling Tools

Control which tools the AI can use:

  1. Open the Tools panel
  2. Toggle tools on/off individually
  3. Disabled tools won't be called by the AI
  4. Preferences are saved per gateway

Why disable tools?

  • Reduce response time (fewer options to consider)
  • Prevent certain actions for safety
  • Focus the AI on specific capabilities

Tool Preferences

Your tool preferences are saved:

  • Per-gateway tool settings are remembered
  • Enable/disable states persist across sessions
  • Preferences sync across devices

Tool Execution

How Tools Are Called

When you make a request:

  1. The AI analyzes your message
  2. It decides which tools (if any) are needed
  3. Tool calls appear in the chat with parameters
  4. Results are shown as tool outputs
  5. The AI uses results to formulate a response

Tool Output Display

Tool results appear in expandable sections:

  • Tool name - Which tool was called
  • Parameters - What inputs were provided
  • Output - The result (expandable for long outputs)
  • Duration - How long execution took

Streaming with Tools

During tool-enabled conversations:

  1. Initial thinking appears as the AI plans
  2. Tool calls show with a loading indicator
  3. Results stream in as they complete
  4. Final response incorporates tool results

Session Management

Remote Chat Sessions

Remote sessions track more than just messages:

  • Tool calls and their results
  • Container state information
  • File changes in workspace
  • Execution history

Session Types

Sessions are marked with their type:

  • Chat - Direct conversations
  • Remote Chat - Sessions with tool access
  • Automation - Script-based sessions

Workspace Persistence

Your workspace files persist across sessions:

  • Files you create remain available
  • Return to previous work
  • Share files across conversations
  • Clean up old files when done

Tool Capabilities

The web search tool provides current information:

User: What are the latest developments in AI?

AI: [Calls web_search with "latest AI developments 2024"]
    Results: [Search results displayed]

Based on recent news, here are the key AI developments...

Tips for web search:

  • Be specific in your questions
  • The AI will formulate good search queries
  • Multiple searches may be used for comprehensive answers

Code Execution

Execute code safely in your container:

User: Calculate the first 10 Fibonacci numbers

AI: [Calls execute_python]
    Code: def fib(n): ...
    Output: [0, 1, 1, 2, 3, 5, 8, 13, 21, 34]

The first 10 Fibonacci numbers are shown above.

Supported languages:

  • Python (with common data science packages)
  • JavaScript/Node.js
  • Bash shell scripts

File Operations

Manage files in your workspace:

User: Save this data to a file and analyze it

AI: [Calls write_file to save data]
    [Calls execute_python to analyze]
    [Calls read_file to show results]

Workspace features:

  • Persistent storage across sessions
  • File size limits apply
  • Organized directory structure

Environment Information

Include Environment Info

Toggle whether to include environment details in remote sessions:

  • Operating system information
  • Available tools and versions
  • Container capabilities
  • Useful for system-specific tasks

Enable this when:

  • Asking about system administration
  • Troubleshooting environment issues
  • Running platform-specific code

Security Considerations

Container Isolation

Your tools run in isolated containers:

  • No access to other users' data
  • Network requests are logged
  • File system is sandboxed
  • Resource limits prevent abuse

What Tools Can Access

Can Access Cannot Access
Your workspace files Other users' files
Public internet Internal networks
Installed packages System resources
Your API keys (for tools) Other credentials

Best Practices

  1. Don't share sensitive data in tool outputs
  2. Review tool calls before sensitive operations
  3. Use specific models for security-critical tasks
  4. Clear workspace when done with sensitive projects

Troubleshooting

Common Issues

Container not starting:

  • Wait a moment and refresh
  • Check gateway status
  • Try a different gateway

Tools not appearing:

  • Ensure remote gateway is selected
  • Check container is initialized
  • Verify tool permissions

Tool execution failing:

  • Check tool parameters
  • Review error messages
  • Try simpler requests first

Error Messages

Error Solution
Container timeout Wait and retry
Tool not found Check tool is enabled
Execution error Review code/parameters
Network error Check gateway connection

Getting Help

If tools aren't working as expected:

  1. Check the request logs for details
  2. Try the same task with a different model
  3. Simplify your request
  4. Contact support if issues persist

Tips for Effective Tool Use

Writing Good Requests

  1. Be specific about what you want
  2. Provide context for complex tasks
  3. Break down large tasks into steps
  4. Specify format for outputs if needed

Combining Tools

The AI can chain multiple tools:

User: Find the current weather and create a Python visualization

AI: [Calls web_search for weather data]
    [Calls execute_python to create chart]
    [Calls write_file to save the image]

Performance Tips

  • Disable unused tools for faster responses
  • Use appropriate models for tool-heavy tasks
  • Start simple and add complexity
  • Check intermediate results for long chains

Next Steps