I like the approach; avoiding a model call just to summarize old context while keeping the cache warm is a nice property.
When it folds away an older chunk of the conversation, does it cause problems if the agent was in the middle of doing something (some multi-turn action) right when that chunk gets folded? Or is it safe to fold at any point regardless of what the agent was doing at that time?
Solid architectural pattern for cost-sensitive long sessions running in one-sessions instead of multiple sessions, and governing into external knowledge, tools, memory and human input audit, btw im in development the tools, ev3lynx727/skeleton-cli, for use case, use skeleton-cli tools to export from backend .db (sqlite) into pre-training datasets !!
We need a new rule in HN where authors of repos with AI generated walls of text as their README may only post their repo if they record themselves reading the readme from top to bottom. Can you do it? Try.
I like the approach; avoiding a model call just to summarize old context while keeping the cache warm is a nice property.
When it folds away an older chunk of the conversation, does it cause problems if the agent was in the middle of doing something (some multi-turn action) right when that chunk gets folded? Or is it safe to fold at any point regardless of what the agent was doing at that time?
Its safe to fold at any point it picks up what it was doing mid edit pretty seamlessly.
Solid architectural pattern for cost-sensitive long sessions running in one-sessions instead of multiple sessions, and governing into external knowledge, tools, memory and human input audit, btw im in development the tools, ev3lynx727/skeleton-cli, for use case, use skeleton-cli tools to export from backend .db (sqlite) into pre-training datasets !!
We need a new rule in HN where authors of repos with AI generated walls of text as their README may only post their repo if they record themselves reading the readme from top to bottom. Can you do it? Try.
This is sick! Great work
[flagged]