workshop-agent

A collection of exercises and demo ai workflows for hackathon

Small Streamlit demo that calls a Dify Workflow to generate a Vacation plan in Markdown and render it in the browser.

This repo contains:

• app.py - Streamlit UI and helper functions that call the Dify Workflows API and extract HTML from the response.
• pyproject.toml - Python dependency manifest.

Features

• Calls Dify Workflows (blocking mode) and shows raw response.
• Extracts raw HTML even when the Workflow returns explanatory text + a Markdown code block (html ... ).
• Renders the HTML inline in Streamlit and offers a download button

Requirements

• Python 3.11+
• See pyproject.toml for pinned dependencies (Streamlit, requests).

Set up Virtual Env (Mac)

python -m venv .venv
source .venv/bin/activate
pip install uv
uv sync

Set up Virtual Env (Windows)

python -m venv .venv
.venv/Scripts/Activate.ps1
pip install uv
uv sync

Check virtual env packages

pip list

Configuration

The app needs the Dify API key and optional configuration. You can provide them via Streamlit secrets or environment variables.

• DIFY_API_KEY (required) — API key for your Dify instance.
• BASE_URL (optional) — Dify base URL (default: https://api.dify.ai).
• WORKFLOW_ID (optional) — workflow id when you prefer the /v1/workflows/{id}/run endpoint.

Recommended: create a file .streamlit/secrets.toml with:

[default]
DIFY_API_KEY = "your_api_key_here"

Run

Start the Streamlit app from the project root:

streamlit run app.py

Open the browser at the address Streamlit prints (usually http://localhost:8501). Fill the form and click "Call Dify for Plan".

Troubleshooting

• If you see Thiếu DIFY_API_KEY in the UI, ensure DIFY_API_KEY is present in .streamlit/secrets.toml or the environment.
• If Streamlit shows the raw response but no HTML, check that your Workflow's final node outputs either html or an output that contains the full HTML. The extractor now checks data.outputs.output first.
• Large HTML payloads may take time; the HTTP timeout in app.py defaults to 180 seconds for blocking response mode (configurable via HTTP_TIMEOUT).

Notes & Next steps

• If you want the app to support more response shapes, add additional candidate paths in extract_html.
• Consider sanitizing or sandboxing untrusted HTML before rendering in production.