Long chat memory and auto-compact

When you would want it

Every model can only read so much at once. Without this, a long chat eventually fails, and a reopened one starts from nothing - so you explain your project again. Compaction keeps the goals, decisions, files and your preferences, and keeps the newest turns word for word.

Where to find it

• All Features → Long chat memory and auto-compact
• Compact now
• View summary
• Turn on or off
• Settings

What you get out of it

• A long conversation never dies at the model's limit - it keeps going for as long as the work does.
• No re-explaining: reopening a chat or switching model keeps what the assistant knew.
• The summary is readable and saved with the conversation, so you can see exactly what the assistant carries forward.
• Long Agents-board runs compact the same way instead of stopping halfway through a big task.

How to use it, step by step

1. Just keep chatting. The ring beside the chat's composer shows how full the model's context is - Context 34%, say - and hovering it gives the numbers.
2. When the conversation reaches 80% of what the model can read, Agentfly asks the same model to summarise the older part. The chat shows Earlier conversation compacted; press View summary to read what it kept.
3. To make room yourself, type /compact and press Enter, or say what matters most: /compact keep the database decisions. Compact now is also on the meter and in the command palette as AI: Compact Conversation.
4. Close Agentfly, come back tomorrow and open the conversation from Conversations. The assistant starts from the saved summary and the messages after it, so it still knows your project, your decisions and what was left to do.
5. Switching model mid-conversation keeps the conversation too. To change when compaction happens - 70%, 80% or 90% - or turn it off, choose Auto-compact settings on the meter.

Why this beats the usual way

Most AI chats either fail when a conversation outgrows the model or quietly drop the oldest messages, and many forget everything when you reopen them. Agentfly summarises instead of dropping, tells you when it has, shows you the summary, and keeps it with the conversation for next time.

Find it in Universal Search

Long chat memory and auto-compact · The assistant · long memory · remember conversation · context · context window · compact · auto compact · summarise · summarize · summary · long chat · conversation too long · forgot · forgets · token limit · prompt too long