Managing Context
Learn how to manage conversation context to keep AI responses fast and accurate.
AI models have a limited "context window" — the amount of conversation they can remember at once. As your conversation grows, managing this context becomes important for maintaining good AI performance.
Understanding Context Usage
The Context Window
Every AI model has a maximum number of tokens (roughly, words) it can process:
- Your messages and attachments use tokens
- AI responses use tokens
- The more context, the slower and more expensive responses become
Context Usage Display
In the chat input area, you'll see a percentage like "15% used":
- Low (under 30%) — Plenty of room, no action needed
- Medium (30-60%) — Consider summarizing if responses slow down
- High (over 60%) — Summarize or clear to improve performance
- Very high (over 80%) — Strongly recommend summarizing
Context limits vary by model. More expensive models typically have larger context windows.
Summarization
Summarization compresses your conversation history into a shorter summary while preserving key information.
How to Summarize
- Click the context usage percentage (e.g., "42% used")
- Select Summarize from the dropdown
- Or type
/summarizein the chat
AI creates a summary of:
- Key decisions made
- Important context about your project
- Current state of the landing page
- Ongoing tasks or requests
When to Summarize
Context Above 50%
High context usage can slow responses. Summarize to free up space.
Slower Responses
If AI is taking longer to respond, context might be too large.
Repeated Instructions
If AI seems to forget earlier decisions, the context may be fragmented.
Changing Direction
When pivoting to a new focus, summarize to consolidate what's done.
Summarization Requirements
- Available only when context usage is 20% or higher
- Cannot summarize while AI is responding
- Creates a new summary each time (replaces previous)
Summarization preserves the important information while discarding verbose back-and-forth. Your page changes are unaffected — only the conversation history is compressed.
Clearing Conversation
Clearing removes all conversation history, starting completely fresh.
How to Clear
- Click the context usage percentage
- Select Clear Conversation from the dropdown
- Or type
/clearin the chat
When to Clear
Starting New Task
When you've finished one task and are starting something unrelated.
Troubleshooting Issues
If AI seems confused or stuck, clearing can reset its understanding.
Context Irrelevant
When previous conversation no longer applies to current work.
Testing Fresh
To see how AI responds without prior context influence.
Clear vs Summarize
| Action | What It Does | Context After | When to Use |
|---|---|---|---|
| Summarize | Compresses history | ~10-20% | Keep working on same project |
| Clear | Removes all history | 0% | Start something new |
Clearing is immediate and permanent. You cannot recover cleared messages. Your landing page code remains unchanged.
Context Management Tips
Break Up Long Sessions
Instead of one marathon session:
- Work on a section (hero, features, etc.)
- Review the results
- Summarize the conversation
- Continue to the next section
This keeps context manageable and responses fast.
Be Concise
Shorter, focused messages use less context:
Uses more context
"I was thinking that maybe we could try to make the button, you know the one in the hero section, a bit more prominent, maybe by changing its color or making it bigger or something like that"
Uses less context
"Make the hero button more prominent with a larger size and brighter color"
Remove Unnecessary Attachments
Large images and documents consume context:
- Only attach what's directly relevant
- Remove attachments before sending if they're not needed
- Use text descriptions when possible instead of images
Choose the Right Model
Different models have different context limits:
- GPT-5 Mini and Gemini 2.5 Flash — Smaller context, good for quick edits
- Claude Sonnet 4.5 and Gemini 2.5 Pro — Larger context, better for complex projects
- Gemini 3 Pro Preview — Largest context windows
If you frequently hit context limits, consider a model with a larger window.
Monitoring Performance
Signs of High Context Usage
- Slower response times
- AI forgetting earlier instructions
- Less coherent or relevant responses
- Error messages about token limits
Signs It's Time to Summarize
- Context above 50%
- Multiple page-length responses in history
- You've changed topics several times
- Session has been going for 30+ minutes