Research using fuzzy search for LLM to find relevant CLI for M365 commands

[waldekmastykarz]

Currently, all CLI for Microsoft 365 commands are exposed to the LLM, which can be inefficient and costly in terms of token usage. This feature request is to research the benefits and feasibility of extending the MCP server with a tool that leverages fuzzy search to help the LLM narrow down relevant commands more efficiently. The goal is to determine if using fuzzy search would reduce the command list the LLM needs to consider and thus save tokens for the user.

Key questions:

• Would a fuzzy search approach improve LLM efficiency for finding commands?
• What are the trade-offs and technical requirements?
• How might this impact user experience and system performance?

Acceptance criteria:

• Analysis of feasibility and efficiency of integrating fuzzy search for command lookup
• Recommendations on next steps for implementation, if beneficial

Please discuss any assumptions, possible approaches, and known limitations in the research.

[Adam-it]

Thanks @waldekmastykarz for the suggestion.
I was thinking of something else to narrow down the initial call that currently gets a simplified JSON of all commands, but I will have a look at fuzzy search 👍.

[Adam-it]

I gave a few libraries a check and
https://www.fusejs.io/
Looks most promising.
Let's try to formulate a implementation plan and must have points for LLM and assign copilot to this