The API serves requests through Bittensor's Subnet 20 to enable tool-based workflows.
The API has a single endpoint for all queries - https://api.gogoagent.ai/v1
OpenAI Chat Completions ExampleBelow are the available parameters for interfacing with the API:
prompt - a simple string with the user query - for use with the /completions endpointExample:
"I need help calculating the tip, what is 10% tip on a bill totalling $100"Example:
[
{"role": "user",
"content": "I need help calculating the tip, what is 10% tip on a bill totalling $100"}
]
Example:
[
{"name": "tip_calculator",
"description": "calculate tip for a total amount",
"arguments": {
"bill_amount": {
"required": True, "type": "number",
"description": "the bill amount in dollars"},
"tip_percent": {
"required": True, "type": "number",
"description": "the tip percentage as a whole number"}
}}
]
Below is a complete Python example demonstrating how to use OpenAI Chat Completions with custom tools. This example includes a function to parse API responses and sample tool functions that can be dynamically executed.
import openai # Import OpenAI SDK for API interaction
# Define API endpoint and model details
MODEL_API = "https://api.gogoagent.ai/v1"
MODEL_NAME = "BitAgent/GoGoAgent"
GOGO_API_KEY = "YOUR_API_KEY_HERE"
# Sample tool: Tip Calculator
def tip_calculator(bill_amount, tip_percent):
return float(bill_amount) * (float(tip_percent) / 100)
# Sample tool: Summation Calculator
def calculator_for_summation(num1, num2):
return float(num1) + float(num2)
# Function to parse the tool call response from GoGoAgent
def parse_tool_call(tool_call):
result = {"name": "", "parameters": {}}
# Remove outer brackets and extract tool name with parameters
tool_call = tool_call.strip("[]") # Strip brackets if present
tool_name, params = tool_call.split("(", 1)
tool_name = tool_name.strip()
params = params.rstrip(")")
# Initialize an empty dictionary for parameters
params_dict = {}
# Extract parameters in key="value" format
for param in params.split(", "):
if "=" in param:
key, value = param.split("=", 1)
value = value.strip('"') # Remove quotes around the value
params_dict[key.strip()] = value
# Store the extracted tool name and parameters
result = {"name": tool_name, "parameters": params_dict}
# Return a list containing the parsed result (supporting future multiple tool calls)
return [result]
# Initialize the OpenAI client with GoGoAgent API credentials
client = openai.OpenAI(
api_key=GOGO_API_KEY,
base_url=MODEL_API
)
# Define the user prompt (requesting help with a tip calculation)
messages = [
{"role": "user",
"content": "I need help calculating the tip; what is 10% tip on a bill totaling $210"}
]
# Define the available tools (functions) for GoGoAgent to select
tools = [
{
"name": "tip_calculator",
"description": "Calculate the tip amount",
"arguments": {
"bill_amount": {
"required": True, "type": "number",
"description": "The bill amount in dollars"
},
"tip_percent": {
"required": True, "type": "number",
"description": "The tip percentage as a whole number"
}
}
},
{
"name": "calculator_for_summation",
"description": "Calculate the sum of two numbers",
"arguments": {
"num1": {
"required": True, "type": "number",
"description": "The first number for summation"
},
"num2": {
"required": True, "type": "number",
"description": "The second number for summation"
}
}
}
]
# Send request to GoGoAgent LLM to determine which tool to use
completion = client.chat.completions.create(
model=MODEL_NAME, # Specify the model name
messages=messages, # Pass user messages
tools=tools, # Provide available tools for LLM to choose from
)
# Extract the tool call response from LLM output
result = completion.choices[0].message.content
# Parse the tool call response into a structured format
tool_calls = parse_tool_call(result)
# Dynamically execute the appropriate function with extracted parameters
tool_name = tool_calls[0]['name']
tool_params = tool_calls[0]['parameters']
result = globals()[tool_name](**tool_params)
# Output the result
print(result)
Here’s a similar example using python and requests:
import json
import requests
# Define the tools
tip_calc_tool = {
"name": "tip_calculator",
"description": "Calculate the tip amount",
"arguments": {
"bill_amount": {
"required": True, "type": "number",
"description": "the bill amount in dollars"
},
"tip_percent": {
"required": True, "type": "number",
"description": "the tip percentage as a whole number"
}
}
}
# Define a summation calculator
sum_calc_tool = {
"name": "sum_calculator",
"description": "Calculate the sum of two numbers",
"arguments": {
"num1": {
"required": True, "type": "number",
"description": "the first number for summation"
},
"num2": {
"required": True, "type": "number",
"description": "the second number for summation"
}
}
}
tools = [tip_calc_tool, sum_calc_tool]
# Setup the messages / prompt
messages = [
{"role": "user",
"content": "I need help calculating the tip; what is 10% tip on a bill totalling $210"}
]
# Send the request to the API
api_endpoint = "https://api.gogoagent.ai/v1/chat/completions"
hdrs = {"Content-Type": "application/json",
"Authorization": "API KEY GOES HERE"}
data = {"messages": messages, "tools": tools}
result = requests.post(api_endpoint, headers=hdrs, json=data)
# print out the results
print(result)
print(result.json())
And for fun, here’s the same query using curl (meaning you can use any libcurl-based library):
curl -X POST https://api.gogoagent.ai/query_subnet \
-H "Content-Type: application/json" \
-H "Authorization: API KEY GOES HERE" \
-d '{
"messages": [
{
"role": "user",
"content": "Need help calculating the tip, what is 10% tip on a bill totalling $100"
}
],
"tools": [
{
"name": "tip_calculator",
"description": "Calculate the tip amount",
"arguments": {
"bill_amount": {
"required": true,
"type": "number",
"description": "the bill amount in dollars"
},
"tip_percent": {
"required": true,
"type": "number",
"description": "the tip percentage as a whole number"
}
}
}
]
}'