Update AI assistant overview

docs/build/ai-assistant.md

@@ -28,7 +28,7 @@ the AI Assistant before it is available. See
 [deployment docs](https://github.com/OpenFn/lightning/blob/main/DEPLOYMENT.md#ai-chat)
 here for help or contact the super user of your instance.
 
-The Assistant is available on app.openfn.org, with usage credits subject your
+The Assistant is available on app.openfn.org, with usage credits subject to your
 project's plan. See [openfn.org/pricing](https://www.openfn.org/pricing) or
 contact [support@openfn.org](mailto:support@openfn.org) for more information on
 paid plans for usage of OpenFn's cloud-hosted platform.
@@ -37,25 +37,24 @@ paid plans for usage of OpenFn's cloud-hosted platform.
 
 ## About the Assistant
 
-The AI Assistant is built on the Claude 3.5 Sonnet model from
-[Anthropic](https://www.anthropic.com/) and is trained with OpenFn documentation
-and example job code.
-
-At the moment, the Assistant is only configured to help with job writing. Later
-versions of the assistant may be rolled out to other pages.
+The AI Assistant is a custom multi-agent system. It has access to OpenFn
+documentation and implementation best practices, so it can answer your questions 
+in the context of the platform.
 
 All chat sessions are shared between all users of the project. You can start a
 new chat session at any time, or open an old one.
 
-Input data and run-time job logs not sent to the model. But we do include your
-step code so that the Assistant can provide a contextually relevant answer.
+You can configure whether your workflow code, run logs and input/output data are
+sent to the model. Sharing this context allows the Assistant to give a more
+contextually relevant answer, but consider carefully whether the data is
+confidential or sensitive before sending it.
 
 ## A Note on Responsible AI Usage
 
-The AI assistant is built on emerging Large Language Model (terminology). Like
-other LLMs and chatbots, its capabilities are impressive, but imperfect.
+The AI Assistant leverages Large Language Models. Like other
+chatbots, its capabilities are impressive, but imperfect.
 
-Remember that ultimately, all responses are generated by an algorithm and YOU,
+Remember that ultimately, all responses are automatically generated and YOU,
 the human in charge, are responsible for how its output is used. You should
 consider all responses critically and verify the output where possible.
 
@@ -64,20 +63,18 @@ consider all responses critically and verify the output where possible.
 
 ## How to access the AI Assistant
 
-The AI Assistant is located in the Inspector, in the Input panel to the left of
-the step code. You may need to open the inspector first - to do this, click on a
-step in the workflow and hit the `</>` button.
+You can access the AI Assistant from either the workflow canvas or an individual
+job step, by clicking on the speech bubble icon in the top right corner.
 
-If there has been a prior chat session on this step, you'll be presented with a
-list of them. Click on one to open up that chat history.
+If there has been a prior chat session, you'll be presented with a list of them.
+Click on one to open up that chat history.
 
-To stat a new session, type a question in the text area at the bottom of the
+To start a new session, type a question in the text area at the bottom of the
 Assistant. Click on the `Send` button to send your question. The Assistant will
-return a respond in a few seconds in the chat interface.
+return a response in the chat interface.
 
 You can close a chat session by clicking the `(X)` button on the top right of
-the chat interface, which will take you back to the list of sessions for this
-step.
+the chat interface, which will take you back to the list of sessions.
 
 :::caution Feedback or Questions about the Assistant?