Appearance
Chat Sessions & Conversation Management
This guide explains how VeriPrompt manages chat sessions, provider selection, and conversation continuity through summarization.
Overview
VeriPrompt's chat system provides:
- Intelligent provider routing with session consistency
- Web search augmentation for real-time queries (weather, news, prices) -- see Web Search
- Conversation summarization for context preservation
- Message limit management with configurable thresholds
Provider Selection
How Providers Are Chosen
When you start a conversation, VeriPrompt selects an AI provider based on several factors:
- Your Selection - If you manually choose a provider in the UI
- Category Settings - Your chat category may have specific routing rules
- Routing Policy - Automatic selection based on performance, cost, or availability
Session Locking
Once a provider is selected for a conversation, all subsequent messages use the same provider/model. This ensures:
- Consistent response style
- Maintained context understanding
- Predictable behavior
Sessions automatically expire after 24 hours of inactivity, allowing the system to select a new optimal provider.
Provider Failover
If your selected provider experiences issues:
- The system automatically retries (up to 3 attempts)
- If needed, it falls back to an alternative provider
- You'll see a notification about the provider change
Message Limits
Limit Tiers
Chat sessions have configurable message limits (counted as question/answer pairs):
| Tier | Limit | Use Case |
|---|---|---|
| Normal | 40 pairs | Standard conversations |
| Medium | 80 pairs | Extended discussions |
| High | 160 pairs | Long-form projects |
Warning Thresholds
As you approach your message limit, you'll see warnings:
| Usage | Banner Color | Message |
|---|---|---|
| 80% | Yellow | "Approaching message limit" |
| 95% | Orange | "Message limit almost reached" + Summarize button |
| 100% | Red | Input blocked - must summarize to continue |
Early Summarization
After 10 message pairs, a "Summarize and start new chat" button appears. This allows you to:
- Save your conversation context
- Start fresh with a clean message count
- Preserve important information
Conversation Summarization
How It Works
When you click "Summarize and Start New Chat":
- Summary Generation - AI creates a structured summary of your conversation
- Context Preservation - Key facts, decisions, and artifacts are preserved
- New Chat Created - A new conversation opens with the summary injected
- Seamless Continuation - The AI remembers your previous discussion
What's Preserved
The summary captures:
- Objective - What you were trying to accomplish
- Key Context - Critical facts and decisions (up to 5 bullets)
- Current State - Where the conversation left off
- Pending Items - Open questions or next steps
- Verbatim Content - Code snippets, URLs, exact values
Rolling Summary Window
VeriPrompt maintains a rolling window of 3 summaries per conversation chain:
- Each summarization creates a new summary
- If you already have 3 summaries, the oldest is automatically removed
- This ensures you always have recent context without unlimited accumulation
Continuing from a Summary
When you continue from a summary:
- The summary is injected into the AI's system prompt
- The AI treats this as its "memory" of your previous conversation
- You can ask follow-up questions naturally
- The AI will reference prior context without you needing to repeat it
Best Practices
For Long Conversations
- Summarize proactively - Don't wait for the limit warning
- Review summaries - Ensure important details are captured
- Use clear topics - Help the AI identify key context
For Complex Projects
- Start with context - Explain your goal at the beginning
- Summarize at milestones - When completing major steps
- Reference artifacts - Include relevant code/URLs that need preservation
For Multiple Sessions
- Use categories - Organize by project or topic
- Continue from summaries - Don't start completely fresh
- Check summary depth - Monitor your continuation chain
Technical Details
Message Counting
- Messages are counted as pairs (user message + assistant response)
- System messages and summaries don't count toward the limit
- Percentage is calculated as:
(pairs / limit) * 100
Summary Token Limit
- Summaries target 400 tokens maximum
- This ensures efficient context injection
- Long conversations are compressed intelligently
Classification Inheritance
When continuing from a summary:
- The new conversation inherits the original classification
- Security policies remain consistent
- Routing rules are preserved
Troubleshooting
"AI says it has no memory"
If the AI doesn't recall prior context:
- Check if you used "Continue from Summary" (not "Start Fresh")
- Verify the summary was successfully created
- The new chat's system prompt should contain the summary
Provider keeps changing
Provider changes can occur due to:
- Session expiration (24 hours)
- Provider outages (automatic failover)
- Category/policy changes
Summary not generating
If summarization fails:
- Ensure the conversation has at least 2 messages
- Check for network connectivity
- Try again - transient provider issues may resolve