The second post in a series about OpenAI’s new “Assistants” API.
We focus here on the Files API, which gives Assistants multi-modal support; I built another Jupyter Notebook to illustrate its usage end-to-end.
See “Using OpenAI’s Assistants API” for Part 1 of the series.
The full Jupyter Notebook is available at my GitHub repository, and shows the step-by-step Python API calls that are required to add an image to an Assistant’s tool resources and then used it in a conversational thread with GPT to solve a problem that wouldn’t be possible using a text-only prompt.
The core is summarized here as:
name = "FrontEnd Dev"
instructions = """You are a dedicated frontend developer,
...
"""
# 0. We need an assistant
new_assistant(name, instructions)
ts_dev = get_asst_by_name(name=name)
#1. Add the uploaded file to its "resources"
client.beta.assistants.update(ts_dev.id,
tool_resources={
"code_interpreter": {
"file_ids":[ui_file.id]
}},
)
ts_dev = get_asst_by_name(name)
print(f"Assistant {ts_dev.name} is now ready, with files {ts_dev.tool_resources.code_interpreter.file_ids}")
the rest of the code is the same as before:
content = """Start creating the base project structure
and main files to implement a UI similar to the one
in the provided `replit-UI.png` screenshot.
"""
# 2. Get a Thread, and append a message to it
thread = new_thread()
add_msg_to_thread(thread, content)
# 4. Get a new Run, and associate it with our Thread
# We will use the Assistant we just created.
run = new_run(thread=thread, asst_name=name)
# 5. We then ask GPT for advice
if wait_on_run(run, thread, timeout=360):
response = get_response(thread)
print(f"{name} says:\n{response}")
else:
print(f"We failed! Status: {run.status} - details: {run.incomplete_details}")
Please see the full Notebook for the complete listing of all the functions.
The full API documentation is here, where there is also a Playground to test with different assistants’ configurations.
Please feel free to reach out if you have an interesting LLM project (OpenAI-related or otherwise) and would like some help in getting it off the ground: my LinkedIn profile is here.
Code Trips & Tips 
Leave a comment