π About
How many times did someone ask you :
"How do you..."
... and how many times did you make the RTFM (Read the Fucking Manual) joke ?
π Well that's all this blog post is all about, we're going to:
-
π Provide a
cli
tool that behaves like all other tools to an AI assistant - π¦₯ Ask it to learn by himself how to use it
- π Make it do the job ... locally !
To achieve this we will use :
-
π€ A core
LLM
engine :gpt-4
- π» A locally running assistant that is able to create an action plan to achieve a goal : Open Interpreter
π― What we'll do
This time, thanks to a custom tool I created last week-end :
β±οΈ Mobitag Go Hackathon 2024-06-22 week-end π€
adriens γ» Jun 23
and ask the AI Assistant to discover the tool and send a custom sms with a custom content to myself with it.
π§βπ« Teach him how to use the tool
π€£ Funniest sms
π Perspectives
The same way we used the --help
pattern for cli
tool, we can use the OpenAPI to tell an AI how to use an API as a tool.
But keep in mind : the better the documentation, the easier and better integration will be... at almost 0 effort (which is our target to scale automation).
Below some examples of how to achieve this on various frameworks & services :
π€ For coders :
β±οΈ < 5'
demo : build & deploy conversational agents (non-coders)
Least but not last, for non coders Google Vertex AI Agent Builder to build and deploy Agents ... within 5 minutes :