Expanded Python instructions for locally running language models

README.md

@@ -155,19 +155,6 @@ In Python, Open Interpreter remembers conversation history. If you want to start
 interpreter.reset()
 ```
 
-### Local chat
-
-```python
-import interpreter
-
-interpreter.local = True #does nothing
-interpreter.model = "openai/model" #Use openai format, model doesn't matter
-interpreter.api_key = "fake_key" #Just needs to have something, doesn't matter
-interpreter.api_base = "http://localhost:1234/v1" #Change to whatever host and port you need. 
-
-interpreter.chat()
-```
-
 ### Save and Restore Chats
 
 `interpreter.chat()` returns a List of messages, which can be used to resume a conversation with `interpreter.messages = messages`:
@@ -212,6 +199,8 @@ interpreter.model = "gpt-3.5-turbo"
 
 ### Running Open Interpreter locally
 
+#### Terminal
+
 Open Interpreter uses [LM Studio](https://lmstudio.ai/) to connect to local language models (experimental).
 
 Simply run `interpreter` in local mode from the command line:
@@ -233,6 +222,21 @@ Once the server is running, you can begin your conversation with Open Interprete
 
 > **Note:** Local mode sets your `context_window` to 3000, and your `max_tokens` to 1000. If your model has different requirements, set these parameters manually (see below).
 
+#### Python
+
+Our Python package gives you more control over each setting. To replicate `--local` and connect to LM Studio, use these settings:
+
+```python
+import interpreter
+
+interpreter.local = True # Disables online features like Open Procedures
+interpreter.model = "openai/x" # Tells OI to send messages in OpenAI's format
+interpreter.api_key = "fake_key" # LiteLLM, which we use to talk to LM Studio, requires this
+interpreter.api_base = "http://localhost:1234/v1" # Point this at any OpenAI compatible server
+
+interpreter.chat()
+```
+
 #### Context Window, Max Tokens
 
 You can modify the `max_tokens` and `context_window` (in tokens) of locally running models.