This is deeply practical, Eddy. We tend to think the quality of an AI collaboration depends primarily on the prompt, but you’re pointing to something more fundamental: the deliberate stewardship of shared context. Knowing when to preserve continuity and when to clear the field may be one of the essential skills of working well with AI. I learned this the hard way too! You're the first person I've heard talk about 'continuation prompts'...and while I've been doing it for a long time, I've never thought to talk about it with anyone else. Cheers!
Thank you. I’ve moved to another solution as of late. It’s now using what is essentially a key value store built into my memory MCP that I call ParaCortex. So now when I close a chat, it will put a continuation summary into the durable cache and I can re-hydrate it in the next chat as needed.
Well that is intriguing...do you mind expanding? Is ParaCortex running locally, and what are you using for the durable store? SQLite, files, Redis, or something else? I’m very interested in building a local version of this for my own long-running conceptual work.
Not local, and simpler than maybe you’d expect. It runs in a free Supabase account, with the MCP itself running as an edge function. So the durable store is just Postgres, no SQLite or Redis in the mix.
The model is four constructs: thoughts, memories, artifacts, and subjects. Thoughts are the working holds, memories are the episodic layer, artifacts are the source records, subjects are the things it all hangs off of. That’s most of it.
Writing up the full build soon in my newsletter, I’ll go into the real detail there.
This is deeply practical, Eddy. We tend to think the quality of an AI collaboration depends primarily on the prompt, but you’re pointing to something more fundamental: the deliberate stewardship of shared context. Knowing when to preserve continuity and when to clear the field may be one of the essential skills of working well with AI. I learned this the hard way too! You're the first person I've heard talk about 'continuation prompts'...and while I've been doing it for a long time, I've never thought to talk about it with anyone else. Cheers!
Thank you. I’ve moved to another solution as of late. It’s now using what is essentially a key value store built into my memory MCP that I call ParaCortex. So now when I close a chat, it will put a continuation summary into the durable cache and I can re-hydrate it in the next chat as needed.
I call this a thought hold and release.
Well that is intriguing...do you mind expanding? Is ParaCortex running locally, and what are you using for the durable store? SQLite, files, Redis, or something else? I’m very interested in building a local version of this for my own long-running conceptual work.
Not local, and simpler than maybe you’d expect. It runs in a free Supabase account, with the MCP itself running as an edge function. So the durable store is just Postgres, no SQLite or Redis in the mix.
The model is four constructs: thoughts, memories, artifacts, and subjects. Thoughts are the working holds, memories are the episodic layer, artifacts are the source records, subjects are the things it all hangs off of. That’s most of it.
Writing up the full build soon in my newsletter, I’ll go into the real detail there.
Amazing! I'll sit tight and wait for the full share! Cheers!
Full build may be over ambitious :)
I’ve rebuilt it 5 times. So I’ll talk a bit conceptually and point folks to the ideas as well as the high level components etc.
Perfect! It's more than I had before.