Conversations & Threads

Understanding how conversations work in FixPanic Chat helps you use the interface more effectively.

Core Concepts

Threads

A Thread represents a single continuous conversation with an agent or group of agents.

Key characteristics:

• Threads preserve context - the agent remembers previous messages
• You can have multiple active threads simultaneously
• Each thread maintains its own history
• Threads are associated with a specific cluster

Context

Context is the accumulated knowledge from your conversation:

• Previous questions and answers
• Commands that were executed
• Results that were returned
• Any files or data shared

Note: Context allows the agent to understand references like "that error" or "the file you showed me" without you repeating everything.

Clusters & Agents

Conversations happen within a context:

• Cluster - The environment (Production, Staging, etc.)
• Agent - The specific agent you're communicating with

When you switch clusters or agents, you're changing who receives your messages.

Managing Conversations

Creating a New Thread

1. Click "New Chat" in the sidebar
2. Select your cluster and agent
3. Start typing your first message

Accessing History

Your past conversations appear in the sidebar:

• Most recent conversations appear at the top
• Click any conversation to resume it
• The full history loads automatically

Searching Conversations

Find specific conversations:

• Use the search box at the top of the sidebar
• Search by keywords in your messages
• Search by agent or cluster name

Deleting Conversations

To remove a conversation:

1. Hover over the conversation in the sidebar
2. Click the delete icon (🗑️)
3. Confirm the deletion

Warning: Deleted conversations cannot be recovered. Make sure you don't need the history before deleting.

Context Management

How Context Works

When you send a message, the agent receives:

1. Your current message
2. Recent conversation history
3. Information about the cluster/agent selected

This context helps the agent:

• Understand what you're referring to
• Remember what commands were already run
• Provide relevant follow-up information

Context Limits

Conversations have context limits:

• Very long conversations may truncate older messages
• The most recent messages are always included
• Important system state is preserved

When to Start Fresh

Consider starting a new thread when:

• Switching to an unrelated topic
• Context from previous messages is no longer relevant
• You want a "clean slate" for troubleshooting
• The conversation has become very long

Preserving Important Information

If you need to remember something across threads:

• Copy important outputs before starting new chats
• Use the clipboard button on code blocks
• Reference specific information in new threads explicitly

Working with Multiple Agents

Switching Agents

You can switch agents within the same thread:

1. Use the agent selector dropdown at the top
2. Select a different agent
3. Continue the conversation

The new agent will see:

• The conversation history
• What the previous agent said
• Commands that were executed

Multi-Agent Coordination

Different agents have different capabilities:

Switch agents based on the task at hand.

Conversation Patterns

Investigative Flow

For troubleshooting issues:

You: "The website is slow"
Agent: [Checks CPU, memory, network]
You: "Focus on the database queries"
Agent: [Analyzes slow queries]
You: "What's the most expensive query?"
Agent: [Shows specific query details]

Exploratory Flow

For understanding a system:

You: "What services are running?"
Agent: [Lists services]
You: "Tell me more about nginx"
Agent: [Shows nginx config and status]
You: "What sites are configured?"
Agent: [Lists virtual hosts]

Command-Driven Flow

When you know what you want:

You: "Run df -h to check disk space"
Agent: [Executes and shows results]
You: "Now show me the largest files in /var/log"
Agent: [Finds and lists large files]

Tips for Better Conversations

Reference Previous Information

You can refer back to earlier parts of the conversation:

• "That error you showed earlier..."
• "For the process you mentioned..."
• "Looking at that output..."

Provide Corrections

If the agent misunderstands:

• "No, I meant the production database"
• "Actually, check the other server"
• "That's not what I was asking about"

Ask for Clarification

When responses are unclear:

• "Can you explain that output?"
• "What does that error mean?"
• "Why would that value be so high?"

Build on Responses

Use the agent's findings:

• "Based on that, what should we do?"
• "Is that related to the memory issue?"
• "Can you fix that automatically?"

Exporting Conversations

To save a conversation:

1. Click the menu icon in the conversation header
2. Select "Export"
3. Choose format (Markdown, Text, or JSON)
4. Download the file

Exports include:

• All messages in the thread
• Timestamps
• Agent/user attribution
• Code blocks (formatted)

Next Steps

• Approvals - Learn about the human-in-the-loop approval system.
• Advanced Features - Explore model selection, deep linking, and more.