i want to build an agent to analyse an

[fringe harbor]

There are ways to build it where context limit is not an issue. The main thing is you don't want to literally read the content of every file in the context.

Most LLMs today will just poke around the structure and a few files and give you a nice report without issue. It depends what kind of analysis you're going for.

If you need something more detailed that does require reading a large number of files you can use sub-agents as a solution. So a "main" agent will look at the structure of the project, and for each component it needs an analysis on, it calls a sub-agent to do the full analysis, gets the report back (along with the report for each component for each call of the sub-agent) and consolidates the report. each agent would have its own context in that case