Skip to main content
This guide explains how to leverage pre-uploaded documents within your automation tasks and workflows by referencing their unique document_id. Prerequisites:
  1. Understanding how to upload documents and obtain their document_id using agent.upload_document as described in Managing Documents.
  2. Familiarity with creating Tasks or defining Workflows.

​
Concept: Referencing document_id in Instructions

The core idea is to embed the document_id (obtained when uploading a file) directly into the instructions you provide for a task or a workflow step. This tells the automation agent to access and utilize the content of that specific document as part of its execution. How it Works (Conceptual): The agent needs to be capable of understanding instructions that reference document IDs. You construct your instructions to guide the agent, for example:
  • “Using the resume (ID: doc_resume_123) and cover letter (ID: doc_cover_456), complete the job application at the target URL.”
  • “Referencing the financial report (ID: doc_q1_report), extract the ‘Total Revenue’ figure and enter it into the spreadsheet.”
  • “Open the contract (ID: doc_contract_abc) and find the section detailing payment terms.”
The exact phrasing depends on the agent’s natural language processing capabilities. The key is providing the document_id as a clear identifier within the context of the instruction, enabling the agent to retrieve and use the correct file.

​
Use Case 1: Submitting Job Applications

Upload a candidate’s resume and cover letter, then use a task or workflow to automatically fill out online job applications using the content from those documents. 
# Assume resume_id = "doc_resume_123" and cover_letter_id = "doc_cover_456"
# were obtained from previous agent.upload_document calls.
# Assume job_application_url = "https://careers.example.com/apply?jobId=XYZ"

try:
    task_details = agent.start_task(
        developer_user_id="candidate_user_1",
        instructions=f"Go to the job application page ({job_application_url}). Fill out the form accurately using the information found in the attached resume (ID: {resume_id}) and cover letter (ID: {cover_letter_id}). Where the application asks for file uploads, use the resume file (ID: {resume_id}) and cover letter file (ID: {cover_letter_id}). Review and submit the application.",
        # target_url is implicitly the job_application_url mentioned in instructions,
        # but could also be set explicitly for clarity.
        # target_url=job_application_url
    )
    if task_details:
        print(f"Job application task started: {task_details.get('task_id')}")
except Exception as e:
    print(f"Error starting job application task: {e}")
Value: Automates the repetitive process of filling application forms, ensuring consistency by using the provided documents as the source of truth.

​
Use Case 2: Financial Data Aggregation (Workflow)

Upload monthly bank statements (e.g., PDFs) and create a workflow to extract key figures, compiling them into a summary. 
# Assume statement_ids = ["doc_jan_report_pdf", "doc_feb_report_pdf", "doc_mar_report_pdf"]
# were obtained from previous agent.upload_document calls.

try:
    tasks = [
        {
            "instructions": f"Access the January bank statement (ID: {statement_ids[0]}). Extract the values for 'Total Deposits' and 'Total Withdrawals'. Record these numbers.",
            # target_url is likely not needed as the agent interacts with the document directly.
        },
        {
            "instructions": f"Access the February bank statement (ID: {statement_ids[1]}). Extract the 'Total Deposits' and 'Total Withdrawals'. Record these numbers.",
        },
         {
            "instructions": f"Access the March bank statement (ID: {statement_ids[2]}). Extract the 'Total Deposits' and 'Total Withdrawals'. Record these numbers.",
        },
        {
            "instructions": "Create a summary of the recorded deposits and withdrawals for the first quarter (January, February, March).",
            # Optional target_url could point to a specific spreadsheet or internal reporting tool.
            # target_url="https://docs.google.com/spreadsheets/d/YOUR_SUMMARY_SHEET_ID"
        }
    ]
    workflow_def = agent.create_workflow(name="Quarterly Financial Summary Extraction", tasks=tasks)
    if workflow_def:
        print(f"Financial summary workflow created: {workflow_def.get('workflow_id')}")

    # You would then typically start this workflow using agent.start_workflow_run(...)

except Exception as e:
    print(f"Error creating financial workflow: {e}")
Value: Automates the extraction of specific data points from multiple structured or semi-structured documents, saving significant manual effort in data entry and aggregation.
This pattern allows for powerful and flexible automations where the agent’s actions are dynamically informed by the content of specific, pre-existing files identified by their document_id.
⌘I