Skip to content
Get packs

Guide · Open source

How to use CowAgent: install, open the Web console and connect models

Updated · written with AI from the project's own documentation and checked against our catalogue data

CowAgent is an open-source personal AI assistant that is meant to run 24/7 on your own computer or server. It can plan multi-step tasks, use tools and skills, keep memory and a knowledge base, and be reached from a web console or chat apps such as WeChat, Feishu, DingTalk, Telegram and Slack. This guide shows you how to install it, open the Web console, connect a model provider and run the basic commands, so you can have a working assistant by the end.

CowAgentzhayujie/CowAgent · ★ 47.8k · MIT

Quick start

  1. 1Check prerequisites. Confirm you have Docker for the Docker option, or Bash or PowerShell for the one-line installers. Have an account ready with a model provider.
  2. 2Run the installer. Run the Bash command on Linux or macOS, or the PowerShell command on Windows. Alternatively, download docker-compose.yml and run docker compose up -d.
  3. 3Open the Web console. Visit http://localhost:9899 in your browser to reach the Web console.
  4. 4Connect a model. Add your model provider in the Web console. No manual file editing is required for providers.
  5. 5Connect a channel. Link a chat app such as WeChat, Feishu, DingTalk, Telegram or Slack if you want to reach the assistant from it.
  6. 6Install a skill. Install a skill from the console or with cow skill install <name>.

What you need before you install CowAgent

CowAgent is self-hosted, so you run it on a personal computer or server that you control. The README lists two prerequisites: Docker, if you use the Docker option, and Bash or PowerShell, if you use the one-line installers.

You also need access to a model provider of your choice, such as Claude, GPT, Gemini or DeepSeek, because the assistant connects to an external LLM rather than bundling one.

  • Docker, for the Docker setup
  • Bash (Linux or macOS) or PowerShell (Windows), for the one-line installers
  • An account with a model provider you want to use

Install CowAgent with the one-line installer or Docker

The one-line installer handles dependencies, configuration and startup. Choose the command for your operating system. On Linux or macOS, run the Bash command. On Windows, run the PowerShell command.

If you prefer containers, download the Docker Compose file and start it in the background. The README gives these as the Docker setup steps.

  • Linux / macOS: run the Bash installer
  • Windows: run the PowerShell installer
  • Docker: download docker-compose.yml, then start it with docker compose up -d
bash <(curl -fsSL https://cdn.link-ai.tech/code/cow/run.sh) irm https://cdn.link-ai.tech/code/cow/run.ps1 | iex curl -O https://cdn.link-ai.tech/code/cow/docker-compose.yml && docker compose up -d

Open the Web console and connect a model

Once CowAgent has started, open http://localhost:9899 in your browser. This is the Web console, where you chat with the Agent, configure models, connect channels and install skills.

Model providers are configured directly in the Web console, so you do not need to edit files by hand. You can route chat, vision, image generation, speech and embeddings to different vendors, and switch providers from the console.

  • Chat with the Agent from the Web console
  • Configure models and providers in the console
  • Connect channels such as WeChat, Feishu, DingTalk, Telegram or Slack
  • Install skills from the console
http://localhost:9899

Run CowAgent on a server safely

To reach the console from outside your machine, the README says to set web_host to 0.0.0.0 in config.json. Set web_password to protect the console, and open port 9899 in your firewall or security group.

Only expose the console when you have set a password, because the assistant can control files, the terminal and the browser.

  • Set web_host to 0.0.0.0 in config.json for external access
  • Set web_password to protect the console
  • Open port 9899 in your firewall or security group

Manage the service with the cow CLI

After installation, the cow command-line tool controls the service. Use it to start, stop and restart CowAgent, check status and logs, update the code, install skills and install browser automation.

  • cow start, cow stop, cow restart: service control
  • cow status, cow logs: status and logs
  • cow update: pull the latest code and restart
  • cow skill install <name>: install a skill
  • cow install-browser: install browser automation
cow start | stop | restart cow status | logs cow update cow skill install <name> cow install-browser

Use the desktop client instead

If you do not want to use a terminal, the README offers a CowAgent Desktop client for macOS and Windows. The backend is bundled, so it is ready to use out of the box. Download it from the project website.

Choose the desktop client for a simpler start, or the installer and Docker options if you plan to run the assistant as a service on a server.

What people use it for

  • A self-hosted personal assistant you message from WeChat, Feishu or Telegram
  • A developer testing an extensible agent framework with skills and MCP support
  • Switching between several LLM providers without rewriting setup
  • A small team experimenting with shared multi-agent conversations

Frequently asked questions

Is CowAgent free?

CowAgent is released under the MIT licence, which is an open-source licence. Your model provider may charge for API use, and the source does not list prices, so check your provider's terms.

What do I need to install CowAgent?

The README lists Docker for the Docker option and Bash or PowerShell for the one-line installers. You also need an account with a model provider such as Claude, GPT, Gemini or DeepSeek.

How do I open the CowAgent web interface?

After it starts, open http://localhost:9899 in your browser. This opens the Web console where you chat, configure models, connect channels and install skills.

Can I run CowAgent on a server?

Yes. The README says to set web_host to 0.0.0.0 and web_password in config.json, then open port 9899 in your firewall. Only do this with a password set.

Is CowAgent safe to use?

It can control files, the terminal and the browser, and connect to external accounts. Only give it access you are comfortable with, and protect the Web console with a password if it is reachable from outside.

How does CowAgent compare with OpenClaw, AutoGPT, Open WebUI or LobeChat?

The source does not compare these projects, so this guide makes no claim about them. CowAgent is a self-hosted assistant with chat-app channels, memory, skills and multi-agent teams. If you only need a basic chatbot, a lighter tool may be a better fit.

New agent skills, repos and AI tools — every week

The best new skills and open-source projects, the fastest-growing repos, and one free prompt. Five minutes to read. Unsubscribe anytime.

Our verdict

CowAgent suits people who want a self-hosted personal assistant reachable from chat apps, and developers who want an extensible agent framework with skills and MCP support. It is a poor choice if you want a hosted service with no setup, or if you only need a basic chatbot, since its feature set is broad and heavier to run.

Install options & downloads