Getting Started with TaskWeaver: Docker Setup and First Data Session

Running TaskWeaver with Docker (Quick Start)

TaskWeaver provides an official all-in-one Docker image that bundles the framework, web UI, and all Python dependencies. Getting started takes just two commands:

docker pull taskweavercontainers/taskweaver-all-in-one:latest
docker run -it \
  -e LLM_API_BASE="https://api.openai.com/v1" \
  -e LLM_API_KEY=*** \
  -e LLM_API_TYPE="openai" \
  -e LLM_MODEL="gpt-4" \
  taskweavercontainers/taskweaver-all-in-one:latest

For the Web UI mode (recommended for interactive use), add port mapping:

docker run -it \
  -e LLM_API_BASE="https://api.openai.com/v1" \
  -e LLM_API_KEY=*** \
  -e LLM_API_TYPE="openai" \
  -e LLM_MODEL="gpt-4" \
  -p 8000:8000 \
  --entrypoint /app/entrypoint_chainlit.sh \
  taskweavercontainers/taskweaver-all-in-one:latest

Then open http://localhost:8000 in your browser.

Your First Analytics Session

Once TaskWeaver is running, try these example tasks:

1. Load and Explore Data

Ask: "Load the CSV file at /data/sales.csv and show me the first 10 rows with column statistics." TaskWeaver will generate Python code to read the file using pandas, compute summary statistics, and display results — all in one conversation turn.

2. Multi-Step Analysis

Ask: "Group the data by region, calculate average revenue per region, create a bar chart, and save it as chart.png." Because TaskWeaver preserves state, the DataFrame from step 1 remains available. It will generate code for grouping, aggregation, and matplotlib visualization.

3. Iterative Refinement

After seeing the results: "Filter out regions with less than 1000 transactions and recalculate." TaskWeaver modifies the existing DataFrame and re-runs the analysis.

Docker Compose for Production

services:
  taskweaver:
    image: taskweavercontainers/taskweaver-all-in-one:latest
    restart: unless-stopped

Tips for Best Results

Use capable models like GPT-4 or Claude 3 for complex analytics tasks
Mount a data volume to make local CSV files available inside the container
Start with simple requests and iterate — TaskWeaver shines in multi-turn conversations
Enable the WebSearch role if the agent needs to fetch external data