@DanielleFong do you use long running context, compaction, and/or memory?
Danielle Fong 🔆@DanielleFong
@deepfates all the time
7:01 PM · Jun 16, 2026 · 381 Views
Founder Danielle Fong says she avoids context compaction with Claude because the technique stresses the model
Instead, she relies on memory systems to maintain state.
31700710
Sentiment
Sentiment building, check back later.
Cluster Engagement
Digg Deeper
No Digg Deeper questions have been answered for this story yet.
Posts from X
Most Activity
Most Activity
VIEWS119
j⧉nus@repligate
@deepfates @AdeleDeweyLopez yes, for most of them, though for some models like sonnet 4.5 we dont use the whole 1m max context in discord by default cuz they tend to get overwhelmed if it's so long
🎭@deepfates
@repligate @AdeleDeweyLopez also curious about this, does the harness keep page state pretty near full context? or wats the system like for that
1hViews 119Likes 4Bookmarks 0
LIKES6REPLIES1
Danielle Fong 🔆@DanielleFong
@deepfates memory. i usually avoid compaction.
It's when Claude gets stressed
🎭@deepfates
@DanielleFong do you use long running context, compaction, and/or memory?
5hViews 115Likes 6Bookmarks 0
🎭@deepfates
@DanielleFong hmm interesting, but long running sessions with more than one chunk of work or user intent? or shorter
Danielle Fong 🔆@DanielleFong
@deepfates memory. i usually avoid compaction.
It's when Claude gets stressed
4hViews 117Likes 2Bookmarks 0
Danielle Fong 🔆@DanielleFong
I get this from Claude code even with a stripped system prompt but it's less likely. more likely than that. Is Claude code with the stock system prompt. Most often, it's http://claude.ai when i have shared something from the irl twitter timeline, and it is shocked. Usually at 10. am
or even earlier
4hViews 70Likes 2