TODO

- Use Jina embedder. 
- get_context should not return the concatenated context, because of unique filtering.

We could increase and decrease the backtick count right? and I would also maybe make a name for the elseif functions like is_opener_ticks and is_closer_ticks. We would need for the outter opener to also allow simply 3 backtick openers without language specification, but any next opener is only an opener if there is lanauge specification otherwise it is a closer, which decreases the nesting level.

#%%

# function llm_context_planner(ctx, tools)
#   response = ai"""
#   What tools do you think you going to need to solve the query:
#   <tools>
#     $tools
#   </tools
#   <query>
#     $ctx
#   </query>
#   """
#   parse_tools(response)
# end

# results = llm_context_planner(ctx=ctx_question, tools=[
#   (:julia_context => "Tool description" => julia_context),
#   (:google_search => "") 
#   ])
#%%
We need to implement the  get_next_msg_contrcutor(::Conversation) and then the ConversationCTX and BetterConversationCTX should just use that.
#%%
Check out the AgeTracker and the Conversation files, and correct the BetterContextCTX tests with the new ideas.
This is the new use of the age things:


#%%
- worktree relative path
- worktree cleanup mistakes...
- merge... accept changes... how? + delete branch?


#%%