update

src/interface.jl

@@ -235,7 +235,7 @@ function chat_mistral_openorca(a::agentReflex)
 end
 
 #WORKING
-function planner_mistral_openorca(a::agentReflex, objective::T) where {T<:AbstractString}
+function planner_mistral_openorca(a::agentReflex)
   """
     general prompt format:
 
@@ -263,6 +263,15 @@ function planner_mistral_openorca(a::agentReflex, objective::T) where {T<:Abstra
         "
   """
 
+  conversation = messagesToString(a.messages)
+  toollines = ""
+  for (toolname, v) in a.tools
+    if toolname ∉ ["chatbox"]
+      toolline = "$toolname: $(v[:description]) $(v[:input]) $(v[:output])\n"
+      toollines *= toolline
+    end
+  end
+  
   assistant_plan_prompt =
     """
     <|im_start|>system
@@ -282,16 +291,21 @@ function planner_mistral_openorca(a::agentReflex, objective::T) where {T<:Abstra
     Use the following format:
     Objective: the objective you intend to do
     Plan: first you should always think about the objective, the info you need and the info you have thoroughly then extract and devise a step by step plan (pay attention to correct numeral calculation and commonsense).
-    p.s.1 each step should be a single action.
-    p.s.2 don't respond to the stimulus yet.
+    p.s.1 each step of the plan should be a single action.
     <|im_end|>
+    $conversation
     <|im_start|>assistant
-    Objective: $objective
-    Plan:
+    Objective: 
     """
-    plan = sendReceivePrompt(a, assistant_plan_prompt, temperature=0.2)
+    result = sendReceivePrompt(a, assistant_plan_prompt, max_tokens=1024, temperature=0.2)
+    @show raw_plan = result
+    x = split(result, "<|im_end|>")[1]
+    x = split(x, "Step")[1]
+    x = split(x, "Plan:")
+    objective = x[1]
+    plan = x[2]
 
-  return plan
+  return objective, plan
 end
 
 
@@ -501,16 +515,16 @@ function conversation(a::agentReflex, usermsg::String; attemptlimit::Int=3)
 
   # a.earlierConversation = conversationSummary(a)
   _ = addNewMessage(a, "user", usermsg)
-  isusetools, objective = isUseTools(a)
+  isusetools = isUseTools(a)
   # newinfo = extractinfo(a, usermsg)
   # a.env = newinfo !== nothing ? updateEnvState(a, newinfo) : a.env
   @show isusetools
 
   if isusetools # use tools before responseing
-    response = work(a, objective)
+    response = work(a)
   end
-  
-  response = chat_mistral_openorca(a)
+
+  response = chat_mistral_openorca(a) #TODO if chatbox is used, skip this 
   response = removeTrailingCharacters(response)
   _ = addNewMessage(a, "assistant", response)
 
@@ -520,17 +534,19 @@ end
 """
   Continuously run llm functions except when llm is getting Answer: or chatbox.
   There are many work() depend on thinking mode.
-""" #WORKING add plan and actor
-function work(a::agentReflex, objective::T) where {T<:AbstractString}
+""" 
+function work(a::agentReflex)
+  actorstate = nothing
   response = nothing
-  error(1)
-  a.memory[:shortterm] = OrderedDict{String, Any}()
-  a.memory[:log] = OrderedDict{String, Any}()
-  a.memory[:shortterm]["user:"] = objective
-  a.memory[:log]["user:"] = objective
-  a.newplan = true
 
   while true  # Work loop
+    #WORKING
+    a.memory[:shortterm] = OrderedDict{String, Any}()
+    a.memory[:log] = OrderedDict{String, Any}()
+
+
+
+
     # plan
     if a.attempt <= a.attemptlimit
       toolname = nothing
@@ -538,17 +554,17 @@ function work(a::agentReflex, objective::T) where {T<:AbstractString}
       if a.newplan == true
         a.attempt += 1
         a.step = 0
-        prompt_plan = planner_mistral_openorca(a)
+        objective, plan = planner_mistral_openorca(a)
         println("")
-        @show prompt_plan
-        response = sendReceivePrompt(a, prompt_plan, max_tokens=1024)
+        @show objective
+        @show plan
 
         # sometimes LLM add not-need word I don't want
-        plan = splittext(response, ["Step 1", "<|im_end|>", "Response", "Execution", 
-                                  "Result", "Recommendation", "My response"])
+        # plan = splittext(response, ["Step 1", "<|im_end|>", "Response", "Execution", 
+        #                           "Result", "Recommendation", "My response"])
         # plan = replace(plan, "Plan:"=>"")
-        println("")
-        @show plan
+        # println("")
+        # @show plan
         
         a.newplan = false
         a.memory[:shortterm]["Plan $(a.attempt):"] = plan
@@ -626,7 +642,7 @@ function work(a::agentReflex, objective::T) where {T<:AbstractString}
 
   # communicates with user
   _ = addNewMessage(a, "assistant", response)
-  return  response
+  return  actorstate, response
 end