diff --git a/API.md b/API.md
index 65cc9fcc..6b2a3745 100644
--- a/API.md
+++ b/API.md
@@ -52,6 +52,8 @@ This creates an assistant with the name `"bar"`, with GPT 3.5 Turbo, with a prom
 Available tools names can be found in the AvailableTools class in backend/packages/gizmo-agent/gizmo_agent/tools.py
 Available llms can be found in GizmoAgentType in backend/packages/gizmo-agent/gizmo_agent/agent_types/__init__.py
 
+Note: If a RAGBot assistant is created (`type` equals `chat_retrieval`), then subsequent API requests/responses for the threads APIs are slightly modified and noted below.
+
 ## Create a thread
 
 We can now create a thread.
@@ -85,7 +87,11 @@ requests.get(
 ).content
 ```
 ```shell
-b'{"messages":[]}'
+b'{"values":[]}'
+```
+For RAGBot:
+```shell
+b'{"values":{"messages":[]}}'
 ```
 
 Let's add a message to the thread!
@@ -102,6 +108,17 @@ requests.post(
     }
 ).content
 ```
+For RAGBot:
+```
+{
+    "values": {
+        "messages": [{
+            "content": "hi! my name is bob",
+            "type": "human",
+        }]
+    }
+}
+```
 
 If we now run the command to see the thread, we can see that there is now a message on that thread
 
@@ -115,6 +132,10 @@ requests.get(
 ```shell
 b'{"values":[{"content":"hi! my name is bob","additional_kwargs":{},"type":"human","example":false}],"next":[]}'
 ```
+For RAGBot:
+```shell
+b'{"values":{"messages":[...]},"next":[]}'
+```
 
 ## Run the assistant on that thread
 
@@ -141,6 +162,10 @@ requests.get('http://127.0.0.1:8100/threads/231dc7f3-33ee-4040-98fe-27f6e2aa8b2b
 ```shell
 b'{"values":[{"content":"hi! my name is bob","additional_kwargs":{},"type":"human","example":false},{"content":"Hello, Bob! How can I assist you today?","additional_kwargs":{"agent":{"return_values":{"output":"Hello, Bob! How can I assist you today?"},"log":"Hello, Bob! How can I assist you today?","type":"AgentFinish"}},"type":"ai","example":false}],"next":[]}'
 ```
+For RAGBot:
+```shell
+b'{"values":{"messages":[...]},"next":[]}'
+```
 
 ## Run the assistant on the thread with new messages
 
@@ -171,6 +196,10 @@ requests.get('http://127.0.0.1:8100/threads/231dc7f3-33ee-4040-98fe-27f6e2aa8b2b
 ```shell
 b'{"values":[{"content":"hi! my name is bob","additional_kwargs":{},"type":"human","example":false},{"content":"Hello, Bob! How can I assist you today?","additional_kwargs":{"agent":{"return_values":{"output":"Hello, Bob! How can I assist you today?"},"log":"Hello, Bob! How can I assist you today?","type":"AgentFinish"}},"type":"ai","example":false},{"content":"whats my name? respond in spanish","additional_kwargs":{},"type":"human","example":false},{"content":"Tu nombre es Bob.","additional_kwargs":{"agent":{"return_values":{"output":"Tu nombre es Bob."},"log":"Tu nombre es Bob.","type":"AgentFinish"}},"type":"ai","example":false}],"next":[]}'
 ```
+For RAGBot:
+```shell
+b'{"values":{"messages":[...]},"next":[]}'
+```
 
 ## Stream
 One thing we can do is stream back responses.
diff --git a/README.md b/README.md
index aeda7426..c1d149f0 100644
--- a/README.md
+++ b/README.md
@@ -301,6 +301,8 @@ This can be a really powerful and flexible architecture. This is probably closes
 these also can be not super reliable, and generally only work with the more performant models (and even then they can 
 mess up). Therefore, we introduced a few simpler architecures.
 
+Assistants are implemented with [LangGraph](https://github.com/langchain-ai/langgraph) `MessageGraph`. A `MessageGraph` is a graph that models its state as a `list` of messages.
+
 **RAGBot**
 
 One of the big use cases of the GPT store is uploading files and giving the bot knowledge of those files. What would it 
@@ -321,6 +323,8 @@ pretty well with a wider variety of models (including lots of open source models
 you don’t NEED the flexibility of an assistant (eg you know users will be looking up information every time) then it 
 can be more focused. And third, compared to the final architecture below it can use external knowledge.
 
+RAGBot is implemented with [LangGraph](https://github.com/langchain-ai/langgraph) `StateGraph`. A `StateGraph` is a generalized graph that can model arbitrary state (i.e. `dict`), not just a `list` of messages.
+
 **ChatBot**
 
 The final architecture is dead simple - just a call to a language model, parameterized by a system message. This allows 
@@ -331,6 +335,8 @@ well.
 
 ![](_static/chatbot.png)
 
+ChatBot is implemented with [LangGraph](https://github.com/langchain-ai/langgraph) `StateGraph`. A `StateGraph` is a generalized graph that can model arbitrary state (i.e. `dict`), not just a `list` of messages.
+
 ### LLMs
 
 You can choose between different LLMs to use.