Fabian Wallwitz
Fabian Wallwitz
@gaborvar Is it fixed? Have you any working example with an additional tool call?
A hybrid solution could balance usability and efficiency: Benefits: Only relevant chunks are retrieved, avoiding data overload. The system dynamically updates context to remain focused on the most important information....
Just do add: Allowing document uploads as any other normal FileStrategy (local or DataLake), where files are stored as originals in blob storage. This approach would enable users to request...
To general decouple the behavior a pluggable tool architecture like MCP would allow to add “archive/brain/memory “ for different datasets like eG personal documents space (blob storage+index/er) for a tool...
But still GPT Vision is then not supported afaik
It’s to slow because all is processing in sequence instead of of parallel . It requires Thread worker pool
The aca container apps log streams says ``` Connecting... 2025-04-13T17:16:27.76906 Connecting to the container 'main'... 2025-04-13T17:16:27.80086 Successfully Connected to container: 'main' [Revision: 'capps-backend-jfnrv5kcsyxqa--zeb9nz5-6d694b59d8-958bc', Replica: 'capps-backend-jfnrv5kcsyxqa--zeb9nz5'] 2025-04-13T17:13:21.6438490Z stderr F 2025/04/13 17:13:21...
I have deployed with the defaults and still get the warning in the portal when viewing the skillset "gptkbindex-skillset" "This skillset contains an AzureOpenAIEmbedding Skill created by previous API versions...
I would be nice to extend the architecture for plug gable tools like MCP does. The adding tools like email, one note, etc could be plugging and by feature flag....
We can use MCP and still Semantic Kernel under the hood – An MCP client is created. – MCP Client is downloading MCP server implementation, in the shape of an...