TaskWeaver is Microsoft’s open-source code-first agent framework for data analytics tasks. It translates user requests into Python code, coordinates plugins as functions, preserves chat and code execution history, supports stateful execution, containers, observability, shared memory, experiences, and vision input for the Planner role.
TaskWeaver is an open-source framework from Microsoft for building code-first AI agents, especially for data analytics and complex task automation. Instead of only passing text between an LLM and tools, TaskWeaver plans through code snippets and coordinates plugins as Python functions.
The key distinction is state. TaskWeaver preserves both chat history and code execution history, including in-memory data such as DataFrames. That makes it more practical for analytical work where intermediate variables, generated charts, fetched datasets, and plugin outputs need to remain available across steps.
Best use cases
Data analytics agents: pull data from databases, transform tables, run anomaly detection, forecast time series, and generate artifacts.
Code-first automation: turn natural-language requests into Python code and plugin calls that can be inspected and debugged.
Domain-specific workflows: wrap proprietary algorithms, databases, APIs, or business rules as plugins.
Stateful multi-step tasks: keep execution history, memory, and intermediate data available during long analytical conversations.
Research and prototyping: compare code-first agent design with LangGraph, AutoGen, OpenHands, CrewAI, and other agent frameworks.
Notable capabilities
Task decomposition, progress tracking, and reflective execution.
Container-mode code execution by default for better isolation.
Vision input support for the Planner role, according to the 2025 official update.
Official screenshot and examples
Official TaskWeaver architecture diagram showing the planner, roles, plugins, and code execution loop.GitHub preview for Microsoft’s TaskWeaver open-source repository.
When to choose TaskWeaver
Choose TaskWeaver when the task is closer to analytical programming than generic chat automation. It is strongest when you need Python, DataFrames, stateful execution, custom plugins, and transparent logs. It is less suitable if you only need a no-code workflow builder or a fully hosted production agent product.
Manus is a hosted general-purpose AI agent that uses cloud VMs, browser automation, files, code and integrations to complete multi-step tasks. This independent guide covers plans and credits, Cloud Browser vs Browser Operator, authenticated actions, privacy, approvals, task design, evaluation and alternatives.
ai-agentfree
2870
Gemini CLI
An open-source AI agent that brings the power of Gemini directly into your terminal.
ai-agentfree
2490
AgentScope
AgentScope is an Apache-2.0 agent framework with ReAct agents, tools, skills, memory, planning, human steering, evaluation, fine-tuning, MCP/A2A integrations, realtime voice, and multi-agent orchestration.
ai-agentfree
3030
Auto-GPT
Auto-GPT is an open-source autonomous-agent project and platform from Significant Gravitas for building, running, and managing AI assistants and workflows.