Guide

AI agents on GitHub: open-source projects and Copilot’s agents

Harrowtech.ai · last checked 13 October 2026

Search for AI agents on GitHub and you find two things: open-source agent projects anyone can download, and the agents GitHub builds into Copilot. This guide covers both. The first half lists five well-known open-source projects, each described from its own README. The second half sets out Copilot’s agent features from GitHub’s documentation.

Harrowtech.ai is independent and not part of GitHub, Microsoft or any project below. For what makes something an agent rather than a fixed workflow, see AI agents vs workflows vs agentic AI.

Open-source AI agent projects on GitHub

GitHub’s ai-agents topic page describes an AI agent as “a software system that autonomously perceives its environment, makes decisions, and takes actions to achieve goals”. On 13 October 2026 the page listed more than 115,000 public repositories under the topic, sorted by most stars. The five below come from the first 20 on that list. The other 15 include plugins and add-ons for existing agents, web scrapers, a prompt guide, a list of skills and more agent frameworks. These five were picked to cover the main kinds: a ready-made agent, a framework for building one, a browser agent, a terminal agent and a course.

Star counts are as GitHub showed them on 13 October 2026 and change daily. A star means someone saved the project, not that they use it.

Hermes Agent (NousResearch/hermes-agent)

The README calls it “the self-improving AI agent built by Nous Research”. It says the agent creates skills from experience, improves them as it uses them, and searches its own past conversations. You can talk to it in the terminal or through Telegram, Discord, Slack, WhatsApp and Signal from one gateway, and it has a built-in scheduler for unattended jobs such as daily reports. It works with the model you choose; the README names Nous Portal, OpenRouter, OpenAI and your own endpoint among them.

Install on Linux, macOS or WSL2:

curl -fsSL https://hermes-agent.nousresearch.com/install.sh | bash

On native Windows the README gives a PowerShell line instead, iex (irm https://hermes-agent.nousresearch.com/install.ps1). Then run hermes to start.

LangChain (langchain-ai/langchain)

The README says: “LangChain is a framework for building agents and LLM-powered applications.” It is a library for developers, not a ready-made agent. For more control over how an agent runs, the README points to LangGraph, and for agents that plan, use subagents and work with files it points to Deep Agents, a package built on LangChain. A separate JavaScript and TypeScript version, LangChain.js, also exists.

uv add langchain

You also need access to a model. The README’s first example calls an OpenAI model.

Browser Use (browser-use/browser-use)

GitHub’s one-line description is “Agents that use the browser.” The README opens with “Navigate the web like a human does” and a demo of booking a driving test. The open-source part is a Python library that runs the agent from your own code; the same company also sells a hosted cloud version. The README says the Python library “is free and MIT-licensed”, while model usage and hosted browsers are charged for separately. It needs Python 3.11 or later.

uv add browser-use

The quickstart then puts an OpenAI API key in a .env file. The README adds that you can use a local browser and a local model through Ollama instead.

Gemini CLI (google-gemini/gemini-cli)

The README calls it “an open-source AI agent that brings the power of Gemini directly into your terminal.” It can query and edit large codebases, run shell commands, work with files and connect to other tools through MCP. Signing in with a personal Google account gives a free tier that the README puts at 60 requests a minute and 1,000 a day.

npm install -g @google/gemini-cli

The README also lists npx @google/gemini-cli to run it without installing, and brew install gemini-cli on macOS or Linux. The guide to Claude Code alternatives sets it beside other coding agents.

AI Agents for Beginners (microsoft/ai-agents-for-beginners)

Microsoft’s repository is a course rather than a tool: “18 Lessons to Get Started Building AI Agents”. The README says each lesson comes with a written lesson, a short video and Python code samples. The lessons run from an introduction and agent design patterns through tool use, planning, multi-agent systems and memory to securing agents. The code uses Microsoft Agent Framework with Microsoft Foundry, which the README says needs an Azure account.

The README suggests forking the repository. Because it holds more than 50 translations, it also gives a lighter clone that skips them:

git clone --filter=blob:none --sparse https://github.com/microsoft/ai-agents-for-beginners.git
cd ai-agents-for-beginners
git sparse-checkout set --no-cone '/*' '!translations' '!translated_images'

Before you run one

• Read the licence file in the repository before building on a project in your business. The table above shows what GitHub listed; the file itself is what counts.
• Expect a model bill. Four of the five sit on top of a model from someone else: the LangChain and Browser Use quickstarts use an OpenAI key, Hermes lets you pick a provider, and Gemini CLI signs in with a Google account. Browser Use’s README states plainly that model usage is charged separately.
• Look before you pipe. The Hermes install line downloads a script and runs it in your shell straight away. Open the script address in a browser and read it first if that is not something you normally do.
• An agent acts with your access. One that can run commands, browse or send messages can do so as you. The guide to whether AI automation is safe covers the risks and safeguards.

The agents GitHub builds into Copilot

GitHub’s own agent features sit inside GitHub Copilot. These are the ones that carry out work rather than suggest text.

Copilot cloud agent

GitHub says it “can research a repository, plan changes, and implement them in the background”, editing files and running tests and linters in “an ephemeral cloud development environment, rather than your local IDE.” You start work from the agents panel, a conversation, or an issue or pull request on GitHub.com. The limits on the same page:

• It changes only the repository named when the task starts, works on one branch at a time and opens one pull request per task.
• Each session runs for at most 59 minutes, a limit that “cannot be extended or bypassed.”
• It works only with repositories hosted on GitHub, and owners or administrators can switch it off for particular repositories.

Agent mode in your editor

GitHub’s chat documentation covers using Copilot Chat and agents “to automate coding tasks by breaking them into steps”. In agent mode Copilot works through a task with tools and corrects its own errors, in your editor with you watching. It is available in VS Code, Visual Studio, JetBrains IDEs, Xcode and Eclipse. In VS Code you choose Agent from the agents dropdown at the bottom of the chat view. Visual Studio needs version 17.14 or later, and an organisation or enterprise administrator may have turned agent mode off.

Copilot CLI

The CLI lets you “use Copilot directly from your terminal”, either as a conversation (start it with copilot) or as a single prompt passed with -p. It runs on Linux, macOS and Windows, from PowerShell or WSL. Its plan mode asks clarifying questions and drafts a plan before writing any code. Cloud sandboxes for it are “in public preview and subject to change.”

Code review and third-party agents

On code review, GitHub says: “Copilot reviews your pull requests, identifies issues, and suggests fixes you can apply in a couple of clicks.” Third-party coding agents let you assign work on GitHub to Anthropic Claude or OpenAI Codex alongside the cloud agent. GitHub says they are “currently in public preview” and must be enabled in your personal, organisation or enterprise policies first. They use GitHub Actions minutes and AI credits, with no extra cost while you stay inside your included amounts.

Which Copilot plan includes what

From GitHub’s plans page, in US dollars. AI credits are used up as you work with the agents.

This guide does not rank or test any of the agents above. GitHub’s pages carry no dates and projects change often, so check the source before you rely on a detail.

Where Harrowtech fits

Harrowtech.ai sells AI setup for businesses, from £720: coding tools set up with the rules and guardrails around how your developers work, AI agents installed, connected to your systems and locked down, and independent testing of an agent you already have, with a written report. It is a service for the setup around tools like these, not a GitHub product. More guides are on the Harrowtech guides page.

Sources

• GitHub, ai-agents topic page (sorted by most stars), read 13 October 2026
• NousResearch/hermes-agent README, read 13 October 2026
• langchain-ai/langchain README, read 13 October 2026
• browser-use/browser-use README, read 13 October 2026
• google-gemini/gemini-cli README, read 13 October 2026
• microsoft/ai-agents-for-beginners README, read 13 October 2026
• GitHub Docs, Copilot agents index, read 13 October 2026
• GitHub Docs, About the Copilot cloud agent, read 13 October 2026
• GitHub Docs, Chat in your IDE (agent mode), read 13 October 2026
• GitHub Docs, About Copilot CLI, read 13 October 2026
• GitHub Docs, About third-party coding agents, read 13 October 2026
• GitHub Docs, Copilot plans, read 13 October 2026

Related: AI agents vs workflows vs agentic AI · Anthropic’s AI agents · Claude Code alternatives · All guides