Skip to content

Installation Instruction and Basic usage

Jump to bottom
Quy Truong edited this page Oct 10, 2025 · 5 revisions

Prerequisties

Install uv for package manager

Windows

powershell -ExecutionPolicy RemoteSigned -c "irm https://astral.sh/uv/install.ps1 | iex"

Linux or MacOS

curl -LsSf https://astral.sh/uv/install.sh | sh

Follow uv instruction for more information here

Installation

Choose either Option 1 or 2 to install agentcrew.

Option 1: Install from PyPI indexer

Linux or Windows

uv tool install --python=3.12 --force agentcrew-ai[cpu] --index https://download.pytorch.org/whl/cpu --index-strategy unsafe-best-match --native-tls

MacOS

uv tool install --python=3.12 --force 'agentcrew-ai[cpu]' --index https://pypi.org/simple/ --index-strategy unsafe-best-match --native-tls

Option 2: Install from the source

Clone the repository

git clone https://github.com/saigontechnology/AgentCrew.git

Use uv to install from source

cd AgentCrew
uv tool install --python=3.12 .[cpu] --reinstall

Configure API keys

After installation completed. you need to configure as least one LLM provider to be able to use agentcrew.

Github copilot authentication

You can authenticate with Github Copilot without manually setup api key by using this command

agentcrew copilot-auth

It will open a browser with device login. Enter the 8 digit showing on the terminal to login.
After login successfully, you can start using agentcrew by running following command:

agentcrew chat

If you would like to use Terminal interface, use the following command instead:

agentcrew chat --console

Other providers

If you don't have github copilot or want to use other LLM providers. you can use API from supported providers.

Run the chat command above when you have not setup any api keys will open the config windows allows you to fill in the keys.

If you already setup LLM providers and want to add more. open config by go to Settings > Global Settings in the GUI or type /edit_config in console UI and fill in the keys.

Web search ability

Agentcrew uses Tavily for search through internet so you need to provide TAVILY_API_KEY to use this feature. go to https://app.tavily.com/home and login with your account to get api key. It has 1000 credit for free per month. Enough for daily use.

Noted: With browser tool, Agent still be able to search using browser but it's ways less effective compare to Tavily.

Setup Agents

The default agent of agentcrew is PromptMaker. You can use this agent to generate system prompt for other agents that focuses to a specific task. Here are some example prompts to create a system prompt agent:

  • System prompt for an Agent specialize on building React project.
Search for latest version of React documentation and best practice when building web application with React, Vite, Tailwind and Shadcn. Make sure all instruction for all libraries match the latest documentation. Use ReAct (Reasoning-Acting) framework to create system prompt for React Agent that's specialized in build web app using React.
  • System prompt for an general Engineer Agent:
Create a system prompt for Engineer Agent that will be helpful for Software Engineer by helping user analyze repo, find solution, identify bug pattern, etc. Use ReAct framework and self-consistency framework to increase agent's critical thinking ability.

Agents config

With each agent, you need to define carefully of these information below to achieve best result. To open agent config, click on Settings > Agents Configuration. In Console UI, use /edit_agent command

Name and Description

Agent's name and description should have clear information of what is the agent's responsibilities. these information will be used for other agents to identify suitable agent for transferring user request. Agent's name must not contains any space.

System prompt

Here is where you define roles, detail responsibilities, instruction details, constraints for agents. Usually, you should use PromptMaker to quickly generate the initial version of your agent and start tuning from there. The more concise and include right enough information, the better agent will behave.

Built-in tools

agentcrew come with bunch of useful built-in tools that use can enable for your agent. Built-in tools are focused for Engineers but they should be suitable for any kind of tasks.

  • memory: Allows agents to use powerful memory system for remembering information across conversations with timeline filtering. This tool also allows agents use adapt tool to learn new behaviors.
  • clipboard: Allows agents to access to system clipboard with read and write permission. Agents can also read image from clipboard if LLM model supported.
  • code_analysis: Allow agents to run analysis through a repository or a folder in repository. The tool's result with contains all of the classes, functions and arguments in each file so agents will have the overall understanding about it. repository must have git inside for this tool to work.
  • web_search: Allows agents searching information on internet and fetch data from an url.
  • image_generation: DEPRECATED this tools needs a re-build for better support newer model and integrate better in the chat flow. Don't use it for now.
  • browser: Allows agents use chrome or chrome-like browsers with a dedicated profiles to browsing and interact with internet. Very useful when you want to do task like booking tickets, checking and posting on social networks.
  • file_editing: Allows agents to edit file on user's machines. Tool has some built-in safety mechanism to prevent agent modify system files and backup in case edit failed. Using search/replace pattern to reduce the token use when edit file.
  • command_execution: Allow agents to run command in user's machines. Command will be executed via bash on Linux/MacOS and powershell on Windows. Tool allows agent to run long-running command and easily to control and interact when needed. There are also safety features that prevent agent run dangerous command.

Noted: All the tools will need user approval to run. for tools like file_editing, command_execution, you should never allow them to run without approved.

Sharing agents with import/export agent features

You can share your agent to your friends or colleagues by using Export/Import features.

To import agents, simply open Agent config by click on Settings > Agents Configuration, then click on Import and choose the agent toml file. In Console UI, you can use /import_agent <file_or_url>. After that use can select it from agent list.

To export agents, open Agent config by click on Settings > Agents Configuration, select agents you want to export (can be multiple), then click on Export, choose the place to save output agent file. In Console UI, you can use /export_agent <agents name separated with comma> <output_file>

Noted: When export agents, MCPs config related to these agents will NOT be exported along with agent. Therefore, when import agents to new machine, you need to re-configure these MCPs again.

Example Agents

Here are some agents that I have created and ready to be use:

1. Engineer Agent:

Agent url: https://agentcrew.dev/agents/Engineer.toml
Good for general engineering tasks.

2. Angular Agent:

Agent url: https://agentcrew.dev/agents/Angular.toml
Very effective in working on Angular project, especially setting up project from scratch. Follow best practices for Angular.

3 Dotnet Agent:

Agent url: https://agentcrew.dev/agents/Dotnet.toml
Specialize in working on asp.net core project. Understand about Clean Architecture with DDD and dotnet cli.

Setup MCPs

Clone this wiki locally