FastAgent
https://github.com/hkuds/fastagent- Category
- AI Agents
- Rank
- No. 1571Tools index
- Pricing
- Open Source
- Type
- TOOL
- Builder
- HKUDS
- GitHub
- 58 stars
- Date
About
Simple, fast, and strong LLM agent framework with tool use, computer use, and orchestration.
What it does
FastAgent turns a natural-language task into coordinated planning, execution, and evaluation work. Specialized agents share task state through an event-driven Kanban model, while grounding backends connect research and reasoning to web, shell, desktop, and MCP tools.
Why it's ranked here
The architecture has substance beyond its headline. It tracks dependencies and task states, retrieves relevant tools, compresses oversized results, and can replan after failed validation. The required local service and operating-system setup make desktop automation less approachable than the core command-line experience.
What's good
The framework separates planning, cross-backend execution, and evaluation into distinct agents. Evaluation can target selected backends or only final actions. Tool schemas are normalized for model compatibility, duplicate tool names receive server-qualified names, and long results are summarized with a truncation fallback. Recording can include conversations, screenshots, and video.
Tradeoffs
Computer control and Python or Bash execution require a separate Flask service. Desktop setup differs across macOS, Linux, and Windows, with extra packages and, on macOS, accessibility and screen-recording permissions. Model API keys are also required. Automatic evaluation and recording add moving parts that users may need to disable or tune.
How to use it well
Use FastAgent for long tasks that mix research with actions across tools or desktop applications, especially when explicit task state, dependency ordering, and selective validation matter. Start with MCP and web workflows, then add the local service when computer control justifies the setup. It does not provide desktop or command execution without that service.
Technical notes+
fastagent/__init__.py maps public attributes to submodules and imports each target on first attribute access through __getattr__, deferring optional or heavy dependency errors until that feature is used. fastagent/__main__.py provides interactive and single-query CLI modes, JSON configuration overrides, workflow and evaluation switches, execution limits, and an ANSI live UI. fastagent/fastagent.py defines configuration precedence, selective evaluation defaults, recording controls, and asynchronous component initialization. fastagent/llm/client.py sanitizes tool schemas, qualifies duplicate tool names with server names, summarizes results above 200,000 characters, and truncates to the same limit if summarization fails. fastagent/kanban/kanban.py keeps system-wide task cards separate from agent-local history in fastagent/memory/memory.py.
Observed
- License
- MIT
- Primary language
- Python 3.10 or newer
- Install surface
- Clone the repository and install dependencies from requirements.txt with pip
- Interfaces
- Python library, interactive CLI, single-query CLI, MCP integration, and local Flask service
- Platform support
- macOS, Linux, and Windows, with platform-specific desktop-control dependencies
Read from README.md, requirements.txt, fastagent/__init__.py, fastagent/__main__.py, fastagent/fastagent.py, fastagent/utils/ui.py, fastagent/llm/client.py, fastagent/agents/base.py, fastagent/llm/__init__.py, fastagent/config/utils.py, fastagent/kanban/enums.py, fastagent/config/loader.py, fastagent/kanban/kanban.py, fastagent/memory/memory.py.
What it can do
Execute tool functions on behalf of users
Natural language requests and available tool definitions → Tool execution results and responses
Interact with computer interfaces automatically
Instructions for desktop/web application tasks → Automated actions performed on computer systems
Orchestrate multi-step workflows
Complex task descriptions requiring multiple operations → Coordinated execution of sequential or parallel tasks
Process and respond to natural language queries
Text prompts and questions → Intelligent text responses and solutions
Integrate and manage multiple AI agents
Agent configurations and task assignments → Coordinated multi-agent system responses
Tags
Tech Stack
Comments (0)
No comments yet
Editorially curated, with community endorsements as a secondary signal. Corrections welcome.