TaskWeaver

open_source

TaskWeaver is Microsoft's open-source, code-first agent framework for planning and executing complex data analytics tasks using LLM-powered agents.

Data & Analytics

AI Agents

AI Frameworks

About

TaskWeaver is a pioneering open-source agent framework developed by Microsoft Research, purpose-built for data analytics automation. Unlike traditional prompt-chaining frameworks, TaskWeaver takes a "code-first" approach: agents generate, validate, and execute code snippets to accomplish complex analytical tasks rather than relying solely on text-based reasoning. This makes it especially powerful for tasks involving data transformation, statistical analysis, and multi-step pipeline execution. The framework supports a rich plugin ecosystem that allows developers to extend its capabilities with custom data connectors, analysis tools, and domain-specific functions. TaskWeaver's multi-agent architecture enables the orchestration of specialized sub-agents, each handling distinct aspects of an analytical workflow — from data ingestion to visualization. An integrated evaluation harness (auto_eval) lets teams benchmark agent performance on custom datasets. Built in Python, TaskWeaver is suitable for data engineers, ML engineers, and researchers who want to automate repetitive analytics workflows or build AI-powered data assistants for their organizations. It includes a playground UI for rapid prototyping and Docker support for containerized deployments. With over 6,100 GitHub stars and contributions from the Microsoft open-source community, TaskWeaver established itself as a foundational reference implementation for code-executing AI agents, despite being archived in early 2026.

Key Features

Code-First Agent Execution: Agents generate and execute real code to accomplish data analytics tasks, producing precise and verifiable results rather than text-only outputs.
Multi-Agent Orchestration: Supports hierarchical multi-agent architectures where specialized sub-agents collaborate to handle different stages of complex analytical workflows.
Extensible Plugin System: Developers can register custom plugins to add domain-specific tools, data connectors, and functions that agents can call during task execution.
Integrated Evaluation Harness: The built-in auto_eval module enables teams to benchmark and evaluate agent performance systematically on custom datasets and task sets.
Playground UI & Docker Support: Comes with a visual playground for rapid prototyping and Docker configurations for reproducible, containerized deployments in production environments.

Use Cases

Automating multi-step data analytics pipelines where an AI agent writes and executes transformation code on raw datasets.
Building enterprise data assistants that allow business analysts to query and analyze large datasets using natural language.
Prototyping and evaluating LLM-powered agents on custom data science benchmarks using the built-in auto_eval harness.
Orchestrating specialized sub-agents to handle distinct stages of a data workflow, such as ingestion, cleaning, analysis, and reporting.
Extending existing Python data stacks with AI-driven automation by registering custom plugins that agents can invoke during task execution.

Pros

Open Source with MIT License: Fully free to use, modify, and deploy, with the permissive MIT license allowing both personal and commercial use without restrictions.
Microsoft-Backed Research Quality: Developed by Microsoft Research with rigorous engineering standards, thorough documentation, and a substantial community of 6,100+ GitHub stars.
Precise Code Execution over Text Reasoning: The code-first approach ensures analytics tasks are executed accurately, making it far more reliable than prompt-only frameworks for data work.
Rich Extensibility: The plugin architecture makes it straightforward to integrate with existing enterprise data stacks, APIs, and custom analysis tooling.

Cons

Repository Archived: Microsoft archived the repository in March 2026, meaning there will be no future bug fixes, security patches, or feature development from the original team.
High Setup Complexity: Requires solid Python development experience and familiarity with LLM APIs to configure, extend, and deploy effectively in production environments.
Narrow Domain Focus: Optimized specifically for data analytics workflows; teams needing general-purpose agent capabilities may find other frameworks more suitable.

Frequently Asked Questions

TaskWeaver is an open-source, code-first agent framework developed by Microsoft Research. It enables AI agents to plan and execute complex data analytics tasks by generating and running actual code, rather than relying purely on text-based reasoning.

Yes. TaskWeaver is released under the MIT open-source license, making it free to use, modify, and distribute for both personal and commercial projects.

No. Microsoft archived the TaskWeaver repository on March 23, 2026, making it read-only. The code remains available and usable, but no new updates, bug fixes, or features will be released by the original team.

TaskWeaver is built in Python. It can be installed via pip and supports standard Python environments as well as Docker containers for production deployments.

TaskWeaver distinguishes itself with its code-first execution model — agents generate and run code to complete tasks rather than composing text or calling pre-defined tools. This makes it especially accurate and auditable for data analytics use cases where computational precision matters.