Python Code Block

Add custom Python logic to your agent with pre-installed libraries. The Python Code Block lets you process data, integrate with external services, perform calculations, and implement custom business logic within your agent’s workflow.

Adding a Python Code Block

Navigate to your agent’s workflow editor

Drag the Python Code Block from the components panel
Connect it to your desired step in the workflow
Use the code editor to implement your custom logic

Access the available variables

Available Variables:

input (str): Input from previous steps.
client_data (dict): Contains user context with the following keys:
- user_id (str): The user’s unique identifier.
- user_name (str): The user’s display name.
- conversation_id (str): The conversation’s unique identifier.
- user_input (str): The user’s input text.
- user_roles (list[str]): The user’s roles.
- user_groups (list[str]): The user’s groups.
- images (list[str]): An array of image URLs.
- additional_info (list[Any]): An array of objects containing additional information passed to the agent.
- step_inputs (list[Any]): The inputs for the current step.
- files (list[str]): The files associated with the request.
- parsed_files (list[dict]): Parsed file content from the Intelligent Parser (available when Include Parsed Files is enabled). Each entry contains:
  - file_id (str): Blob storage identifier.
  - name (str): Original file name.
  - content (str): Parsed text output (markdown or structured content).
  - content_type (str): MIME type of the original file.
- variables (dict[str, str]): Custom variables configured in the agent.
execution_parameters (dict): Contains the following keys:
- execution_id (str): The unique identifier for the execution.
- step_results_by_type (dict): A dictionary containing the results of the previous steps, ordered by completion, with the following keys:
  - model (list[Any]): The results of the model steps.
  - data_source (list[Any]): The results of the data search steps.
  - memory_load (list[Any]): The results of the memory load steps.
  - memory_store (list[Any]): The results of the memory store steps.
  - python (list[Any]): The results of the python steps.
  - router (list[Any]): The results of the router steps.
  - tool_action (list[Any]): The results of the tool action steps.
  - agent (list[Any]): The results of the agent steps.

Accessing Parsed File Content

Python code blocks can access the content of files processed by the Intelligent Parser earlier in your agent pipeline. This allows you to programmatically manipulate, transform, or extract data from uploaded documents (PDFs, spreadsheets, etc.) directly in Python, without routing them through an LLM step.

Enable Include Parsed Files

The feature is opt-in because file payloads can be large. To enable it:

Open your Python code block in the Agent Studio.
Toggle “Include Parsed Files” in the code block toolbar.

The toggle is available in both the classic editor and the vibe experience.

Access Parsed Files in Your Code

When enabled, parsed file content is available in client_data["parsed_files"] as a list of objects. Each entry contains:

Field	Description	Example
`name`	Original file name	`"report.pdf"`
`content`	Parsed text output (markdown or structured content)	`"# Report\n\nData..."`
`file_id`	Blob storage identifier	`"a1b2c3d4-e5f6-7890-abcd-ef1234567890"`
`content_type`	MIME type of the original file	`"application/pdf"`

for file in client_data.get("parsed_files", []):
    print(file["name"])         # e.g. "report.pdf"
    print(file["content"])      # Parsed markdown/structured output
    print(file["file_id"])      # Blob storage identifier
    print(file["content_type"]) # e.g. "application/pdf"

Example: Extract and Process Parsed Documents

import json

results = []
for file in client_data.get("parsed_files", []):
    if file["content_type"] == "application/pdf":
        results.append({
            "source": file["name"],
            "data": file["content"]
        })

output = json.dumps(results)

Parsed Files vs. LLM Step Attachments

If your pipeline also uses LLM steps with the Include Attachments option, here is how the two approaches compare:

	Python Code Block (Include Parsed Files)	LLM Step (Include Attachments)
Content delivery	Full content in a single entry per file	Chunked into ~8 KB pieces
Processing	Programmatic (your Python code)	Model-driven (prompt-based)
Best for	Data extraction, transformation, filtering	Summarization, Q&A, analysis

Use Include Parsed Files when you need full programmatic control over the document content. Use Include Attachments on an LLM step when you want the model to reason over the content.

Access the available libraries

airia==0.1.39aiohappyeyeballs==2.6.1aiohttp==3.13.4aiosignal==1.4.0annotated-types==0.7.0anyio==4.12.1asgiref==3.8.1asn1crypto==1.5.1atpublic==7.0.0attrs==25.3.0azure-ai-vision-imageanalysis==1.0.0azure-core==1.38.0azure-cli==2.83.0azure-identity==1.25.1azure-storage-blob==12.28.0b1azure-keyvault==4.2.0beautifulsoup4==4.13.4boto3==1.42.30botocore==1.42.30Brotli==1.2.0cachetools==7.0.0certifi==2025.4.26cffi==2.0.0charset-normalizer==3.4.1click==8.1.8cloudpickle==3.0.0cmdstanpy==1.3.0contourpy==1.3.2cryptography==46.0.7cycler==0.12.1Cython==3.0.12dataclasses-json==0.6.7decorator==5.2.1defusedxml==0.7.1Deprecated==1.3.1distro==1.9.0duckdb==1.4.4dnspython==2.7.0email_validator==2.2.0et_xmlfile==2.0.0fastapi==0.129.0fastapi-cli==0.0.20filelock==3.20.3filetype==1.2.0fonttools==4.61.0frozendict==2.4.6frozenlist==1.6.0gevent==25.4.2geventhttpclient==2.3.3googleapis-common-protos==1.70.0google-api-python-client==2.189.0greenlet==3.2.1grpcio==1.71.0gurobipy==12.0.2h11==0.16.0holidays==0.71html5lib==1.1httpcore==1.0.9httptools==0.7.1httpx==0.28.1idna==3.11imageio==2.37.2imageio-ffmpeg==0.6.0importlib_metadata==8.6.1importlib_resources==6.5.2iniconfig==2.1.0isodate==0.7.2Jinja2==3.1.6jiter==0.12.0jmespath==1.0.1joblib==1.5.3json_repair==0.43.0jsonpatch==1.33jsonpointer==3.0.0kiwisolver==1.4.8langchain-core==1.2.28langchain-openai==1.0.2langchain-text-splitters==1.0.0langsmith==0.7.31lxml==6.1.0markdown-it-py==4.0.0MarkupSafe==3.0.2marshmallow==3.26.2matplotlib==3.10.1mdurl==0.1.2moviepy @ git+https://github.com/Zulko/moviepy@c5b2423 # pinned to commit with Pillow 11.2+ supportmultidict==6.4.3multitasking==0.0.12mypy_extensions==1.1.0nltk==3.9.4numpy==1.26.4docxtpl==0.20.2openai==2.7.1opencv-python==4.11.0.86opencv-python-headless==4.11.0.86openpyxl==3.1.5opentelemetry-api==1.39.1opentelemetry-exporter-otlp==1.39.1opentelemetry-exporter-otlp-proto-common==1.39.1opentelemetry-exporter-otlp-proto-grpc==1.39.1opentelemetry-exporter-otlp-proto-http==1.39.1opentelemetry-instrumentation==0.60b1opentelemetry-instrumentation-asgi==0.60b1opentelemetry-instrumentation-fastapi==0.60b1opentelemetry-proto==1.39.1opentelemetry-sdk==1.39.1opentelemetry-semantic-conventions==0.60b1opentelemetry-util-http==0.60b1orjson==3.11.7packaging==24.2pandas==2.2.3pdf2image==1.17.0peewee==3.18.2pgeocode==0.5.0pillow==12.2.0pillow-avif-plugin==1.5.2pip-upgrade==0.0.6platformdirs==4.5.1playwright==1.58.0pluggy==1.5.0proglog==0.1.11prometheus_client==0.21.1propcache==0.3.1prophet==1.1.6protobuf==6.33.5psycopg2-binary==2.9.11pybase64==1.4.2pycparser==2.22pydantic==2.11.3pydantic-settings==2.9.1pydantic_core==2.33.1pyee==13.0.0Pygments==2.20.0PyJWT==2.12.0PyMuPDF==1.27.1pymupdf4llm==0.3.4PyMySQL==1.1.2pyodbc==5.3.0pyOpenSSL==26.0.0pyparsing==3.3.1pytesseract==0.3.13pytest==9.0.3pytest-asyncio==1.3.0python-dateutil==2.9.0.post0python-docx==1.1.2python-docx-replace==0.4.4python-dotenv==1.2.2python-multipart==0.0.26python-pptx==1.0.2python-rapidjson==1.20pytz==2025.2PyYAML==6.0.2regex==2025.10.23requests==2.33.0requests-toolbelt==1.0.0rich==14.3.3rich-toolkit==0.14.8s3transfer==0.16.0scipy==1.15.2shellingham==1.5.4simple-salesforce==1.12.6slack-sdk==3.35.0six==1.17.0sniffio==1.3.1snowflake-connector-python==4.4.0snowflake-snowpark-python==1.45.0snowflake.core==1.4.0sortedcontainers==2.4.0soupsieve==2.8.3SQLAlchemy==2.0.40stanio==0.5.1starlette==0.52.1tenacity==9.1.2tomlkit==0.13.2tqdm==4.67.1tritonclient==2.56.0typer==0.21.1typing-inspect==0.9.0typing-inspection==0.4.2typing_extensions==4.13.2tzdata==2025.3tzlocal==5.3.1ujson==5.12.0urllib3==2.6.3uvicorn==0.34.2uvloop==0.21.0watchfiles==1.0.5weaviate-client==4.16.9webencodings==0.5.1websockets==15.0.1wrapt==1.17.2XlsxWriter==3.2.9yarl==1.22.0yfinance==0.2.57youtube-transcript-api==1.0.3zipp==3.23.0zope.event==6.1zope.interface==7.2zstandard==0.23.0

Overview

Agent Basics

Interface Options

Sharing Agents

Adding a Python Code Block

Accessing Parsed File Content

Enable Include Parsed Files

Access Parsed Files in Your Code

Example: Extract and Process Parsed Documents

Parsed Files vs. LLM Step Attachments

Overview

Agent Basics

Interface Options

Sharing Agents

Documentation Index

​Adding a Python Code Block

​Accessing Parsed File Content

​Enable Include Parsed Files

​Access Parsed Files in Your Code

​Example: Extract and Process Parsed Documents

​Parsed Files vs. LLM Step Attachments

Adding a Python Code Block

Accessing Parsed File Content

Enable Include Parsed Files

Access Parsed Files in Your Code

Example: Extract and Process Parsed Documents

Parsed Files vs. LLM Step Attachments