update ConversationMemory.md (and related)

Writerside/s.tree

@@ -8,15 +8,20 @@
     <toc-element topic="Overview.md">
         <toc-element topic="ProjectNotes.md">
             <toc-element topic="Chatbot.md">
+                <toc-element topic="User-Interface.md"/>
                 <toc-element topic="AI.md">
-                    <toc-element topic="Tools.md"/>
-                    <toc-element topic="LangChain.md">
+                    <toc-element topic="Tools.md">
+                        <toc-element topic="Foam.md"/>
+                        <toc-element topic="ContinueDev.md"/>
                         <toc-element topic="OpenAI.md"/>
+                        <toc-element topic="LangChain.md">
+                        </toc-element>
                     </toc-element>
                     <toc-element topic="Audio.md">
                         <toc-element topic="Whisper.md"/>
                     </toc-element>
                     <toc-element topic="Language.md">
+                        <toc-element topic="Retrieval-Augmented-Generation.md"/>
                         <toc-element topic="Agent.md">
                             <toc-element topic="LLMChain.md">
                                 <toc-element topic="LargeLanguageModel-LLM.md">
@@ -37,8 +42,6 @@
                     </toc-element>
                 </toc-element>
                 <toc-element topic="FileSystem.md">
-                    <toc-element topic="Foam.md"/>
-                    <toc-element topic="ContinueDev.md"/>
                 </toc-element>
                 <toc-element topic="TimedEvents.md"/>
             </toc-element>

Writerside/topics/ConversationMemory.md

@@ -12,4 +12,11 @@ Future development:
 [ ]- Token Buffer (keep a buffer of messages that is less than `n` tokens)
 [ ]- Summary Buffer (keep a buffer of messages that is less than `n` tokens, but also summarize the messages)
 [ ]- Time weighted Vectorstore (keep a vectorstore of messages, weighted by time)
+    - I would want to implement something like this, but implement something like:
+        - All messages tied to the [User](User.md) are in the vector store
+        - Retriever accesses this vector store based on similarity to current [Couplet](Couplet.md) [Context](ContextRoute.md)inversely weighted by time (recent messages advantaged)
+        - Messages within [Chat](Chat.md) context given a (very?) big advantage over messages outside of context (Tunable via [UserInterface](User-Interface.md)) 
+        - Give a (very strong) advantage to 'in-chat' (ensuring they will be selected in the [RAG](Retrieval-Augmented-Generation.md) model), 
+
+or like, sigmoid or whatever so that recent messages within the conversation are weighted more heavily than older messages.
     - [https://js.langchain.com/docs/modules/data_connection/retrievers/time_weighted_vectorstore](https://js.langchain.com/docs/modules/data_connection/retrievers/time_weighted_vectorstore)

Writerside/topics/Foam.md

@@ -1,3 +1,5 @@
 # Foam
 
-A way to make [VsCode](VsCode.md) interact with Markdown like [Obsidian](Obsidian.md) does
+A way to make [VsCode](VsCode.md) interact with Markdown like [Obsidian](Obsidian.md) does
+
+Not technically an AI tool, but its well suited to become one (like [Obsidian](Obsidian.md) is)

Writerside/topics/Retrieval-Augmented-Generation.md

@@ -0,0 +1,3 @@
+# Retrieval Augmented Generation (RAG)
+
+https://python.langchain.com/docs/use_cases/question_answering/

Writerside/topics/User-Interface.md

@@ -0,0 +1,23 @@
+# User Interface
+
+The stuff we want to let the [User](User.md) control about the [ChatBot](ChatBot.md)
+
+## LLM Config
+- [ ] ModelName (default = "gpt4-1106-preview" for now, "gpt4-turbo" when it becomes available)
+- [ ] Temperature (default = 0.7 I think)
+- [ ] **OtherModelParameters
+
+## Conversation Memory
+- [ ] Type, window, etc
+
+## [Vectorstore](VectorStore.md) config
+- [ ] Which [Vectorstore](VectorStore.md) [contexts](ContextRoute.md) to make available:
+  - [ ] [FileSystem](FileSystem.md) interface kinda thing)
+  - [ ] Individual Documents (think like textbooks, websites, etc)
+
+- [ ] Retrieval
+  - [ ] By Tag (human applied)
+  - [ ] By Topic (AI extracted)
+  - [ ] By Summary (AI generated)
+    - Generate summaries at multiple scales, like I did with the [PaperSummary](https://github.com/jonmatthis/chatbot/blob/main/chatbot/ai/workers/green_check_handler/parse_green_check_messages.py#L20) thing in OG [ClassBot](ClassBot.md) 
+

Writerside/topics/VectorStore.md

@@ -2,12 +2,13 @@
   - [ ] Ingestion Method
     - [ ] Ingestion interface
 	    - [ ] [FileSystem](FileSystem.md)
-- [ ] Embedding method #Langchain
-      - [ ] #OpenAI
-      - [ ] #HuggingFace
+- [ ] Embedding method 
+      - [ ] [OpenAI](OpenAI.md)
+      - [ ] HuggingFace
       - [ ] Local
     - [ ] Persistence Method
-      - local #Chroma
+      - local 
+        - [Chroma](Chroma.md)
       - hosted #MongoDB?
     - [ ] Retriever Method ?
   - [ ] Sources