Can you say more? This sounds like it could be more useful than trying to build a knowledge base (say qmd) and then get the models to search it for the right context each time.
This is in fact what we do (with higher order abstractions now built on top of this). This builds self evolving interactive knowledge base and puts it into a QMD searchable index. The indexer is already open source: https://github.com/jibs/duffel
Comments
what do you mean by this ? do you rewrite the context in your proxy ?
I imagine they're capturing all the tokens+context from pi sessions from every developer.
Yeah.
Can you say more? This sounds like it could be more useful than trying to build a knowledge base (say qmd) and then get the models to search it for the right context each time.
This is in fact what we do (with higher order abstractions now built on top of this). This builds self evolving interactive knowledge base and puts it into a QMD searchable index. The indexer is already open source: https://github.com/jibs/duffel