How to get response_format working with assistants #586
Description
How do I force the assistant to respond in json ? This is my code. Whatever I do the response always comes back with the ```json prefix
`# == Service: KpiXlExtractionEngine
Processes an MIS Excel file and extracts KPI records per tenant and company
class KpiXlExtractionEngine
def initialize(file_path:, entity_id:, portfolio_company_id:)
@file_path = file_path
@entity_id = entity_id
@portfolio_company_id = portfolio_company_id
end
Main entry point to extract KPI records
def process
# Set your API key
client = OpenAI::Client.new(access_token: Rails.application.credentials["OPENAI_API_KEY"],)
# === 1. Upload the Excel File ===
file_upload_response = client.files.upload(
parameters: {
file: Faraday::UploadIO.new("tmp/MIS sample.xlsx", "application/vnd.openxmlformats-officedocument.spreadsheetml.sheet"),
purpose: "assistants"
}
)
file_id = file_upload_response.dig("id")
puts "Uploaded file ID: #{file_id}"
# === 2. Create an Assistant (only once, reuse the assistant_id if already created) ===
assistant = client.assistants.create(
parameters: {
name: "Excel KPI Extractor #{@entity_id} #{@portfolio_company_id} #{Time.now}",
instructions: "You are an financial analyst who can scan and review Excel files. You will be provided with an Excel file containing financial data. Your task is to extract metrics in all the worksheets in the file, such as but not limited to Revenue, GMV etc for all avaliable dates. Please provide the extracted values across all dates for a specific KPI in json format for example {revenue: {date: 10000, date: 20000}}. Always return pure json.",
response_format: { type: "json_object" },
tools: [{ type: "code_interpreter" }],
tool_resources: {
code_interpreter: {
file_ids: [file_id] # See Files section above for how to upload files
}
},
model: "gpt-4o"
}
)
assistant_id = assistant["id"]
puts "Assistant ID: #{assistant_id}"
# === 3. Create a Thread ===
thread = client.threads.create
thread_id = thread["id"]
# === 4. Create a Message with the File ===
client.messages.create(
thread_id: thread_id,
parameters: {
role: "user",
content: "Please extract the Revenue, Net Current Assets, and Number of Distributors from this Excel file.",
}
)
# === 5. Run the Assistant ===
run_id = client.runs.create(
thread_id: thread_id,
parameters: {
assistant_id: assistant_id,
response_format: { type: "json_object" },
}
)["id"]
puts "Run ID: #{run_id}"
# === 6. Poll for Completion ===
while true do
response = client.runs.retrieve(id: run_id, thread_id: thread_id)
status = response['status']
case status
when 'queued', 'in_progress', 'cancelling'
print "."
sleep 3 # Wait one second and poll again
when 'completed'
break # Exit loop and report result to user
when 'requires_action'
# Handle tool calls (see below)
when 'cancelled', 'failed', 'expired'
puts response['last_error'].inspect
break # or `exit`
else
puts "Unknown status response: #{status}"
end
end
# === 7. Get the Response ===
messages = client.messages.list(thread_id: thread_id)
response_text = messages["data"].first["content"].map { |c| c["text"]["value"] }.join("\n")
puts "\n==== Extracted Metrics ===="
puts response_text
end
end
`