Skip to content
Agentdex

Blog · · Agentdex Editorial

GitHub Copilot Agents (2026): Agent Mode, Cloud Agent, Custom Agents and CLI

"GitHub Copilot (Agentdex profile) agents" now means several different things. There is agent mode in your editor, a cloud agent that works on GitHub and opens pull requests, the Copilot CLI with its own built-in agents, and custom agents you define in Markdown. This guide explains what each one does, where it runs, what it costs in AI Credits and where its limits are. Everything here was checked against GitHub's documentation and plans page on October 24, 2026.

Key Takeaways

  • Agent mode runs inside your IDE and edits files locally while you watch. The cloud agent runs on GitHub in a temporary environment and hands you a branch or pull request.
  • Custom agents are Markdown files with YAML frontmatter. They give Copilot a specific role, instructions and a limited tool set, and they can be shared across a repository, organization or enterprise.
  • The Copilot CLI ships with built-in agents such as explore, task, code-review, research and security-review.
  • Agent work uses AI Credits. Code completions do not. Free includes limited agent use; Pro is $10/month with 1,500 credits.
  • The cloud agent has hard limits: one repository, one branch and one pull request per task, and a 59-minute maximum session.

The four kinds of Copilot agent

AgentWhere it runsWhat you get backBest for
Agent modeYour IDE (VS Code, Visual Studio, JetBrains and others)Edits in your working copyInteractive, multi-file changes you want to steer
Cloud agentA temporary GitHub-hosted environmentA branch and, if you want, a pull requestWell-scoped issues you can hand off and review later
Copilot CLIYour terminalEdits, commands and answers in the terminalTerminal-first developers and scripted workflows
Custom agentsThe cloud agent, CLI and supported IDEsWhatever the agent's profile allowsRepeatable roles such as a test writer or a planner

Agent mode in your editor

Agent mode is the hands-on option. You describe a task in Copilot Chat, and Copilot plans the change, edits several files, can run terminal commands and responds to errors, all inside your local workspace. You see every change before you keep it, so it suits work where you want to stay involved: refactors, new features in a codebase you know, or fixing a failing test run.

GitHub Copilot is available in VS Code, Visual Studio, JetBrains IDEs, Xcode, Vim/Neovim, Zed (Agentdex profile) and Eclipse, among others. Feature support varies by editor, and VS Code usually gets new agent features first. The trade-off is simple: agent mode is fast and visible, but it ties up your editor and your attention while it works.

Copilot cloud agent

The cloud agent (older pages still call it the Copilot coding agent) works independently in a temporary development environment powered by GitHub Actions. It can research the repository, make a plan, change code, and run your tests and linters before it hands the work back.

You can start a task from the agents panel on GitHub, from a Copilot Chat conversation, or by assigning an issue to Copilot. Copilot creates the branch, writes commit messages and pushes the changes. You can review the diff first and then ask for a pull request, or ask for one straight away. Every session has a log so you can see what the agent did and why, and you can steer it with follow-up comments.

GitHub documents some firm limits:

  • It works on one repository per task and makes changes in one branch, with one pull request per task.
  • A session can run for up to 59 minutes. You can set a shorter limit with timeout-minutes in copilot-setup-steps.yml.
  • It only works with repositories hosted on GitHub.
  • Repository rulesets can block it. Admins can add Copilot as a bypass actor, and repository admins can turn the agent off entirely.

Two related features are worth knowing. Automations let you run the cloud agent on a schedule or in response to repository events, set up from the Agents tab in eligible private and internal repositories. GitHub Agentic Workflows let you describe automation in Markdown files that run as GitHub Actions. GitHub's plans page also lists third-party coding agents, including Claude and Codex (Agentdex profile), in preview.

Custom agents

A custom agent is an agent profile: a Markdown file with YAML frontmatter, usually stored in your repository's .github/agents folder. The frontmatter sets a description (required), and optionally a name, a model, a tools list and, for the cloud agent, mcp-servers. The Markdown body holds the instructions, up to 30,000 characters.

Two practical uses stand out:

  • A test specialist that writes and improves tests but is told not to touch production code.
  • An implementation planner limited to read, search and edit tools, so it writes plans rather than running commands.

Setting tools: [] disables all tools, and leaving tools out gives the agent everything available. Profiles can live at repository, organization or enterprise level. When names clash, the lowest level wins, so a repository agent overrides an organization agent with the same name. The Copilot CLI also supports personal profiles.

Custom agents work in the cloud agent, the Copilot CLI, VS Code, Visual Studio, JetBrains IDEs, Eclipse and Xcode. JetBrains, Eclipse and Xcode support is in public preview.

Copilot CLI and its built-in agents

The Copilot CLI brings the agent into your terminal. Besides your own custom agents, it includes built-in agents for common jobs: explore, task, general-purpose, code-review, research, rubber-duck and security-review. The Free plan includes limited CLI use, so you can try it before paying. If you already work in the terminal with tools like Claude Code, the CLI is the most natural way in.

What Copilot agents cost

Since GitHub's move to AI Credits, agent features draw from a monthly credit allowance (1 AI Credit = $0.01). Code completions and next edit suggestions are unlimited on paid plans and do not use credits.

PlanPrice (Oct 2026)Monthly AI Credits
Free$0Limited chat and agent use; 2,000 completions a month
Pro$10/month1,500
Pro+$39/month7,000
Max$100/month20,000
Business$19/seat/month1,900 per seat, pooled
Enterprise$39/seat/month3,900 per seat, pooled

Credit use per task depends on the model and the length of the session, so a long cloud-agent run costs more than a quick chat. Start on a lower tier, watch your usage for a couple of weeks, then decide. For the full breakdown, see our GitHub Copilot pricing guide.

Watch-outs

  • Review everything. Agents can produce code that compiles but is wrong or insecure. Treat the output like a pull request from a new teammate.
  • Scope tasks tightly. The cloud agent does best with a clear issue, acceptance criteria and working tests. Vague tasks burn credits and time.
  • Check data settings. On individual plans, GitHub may use your data to improve models unless you opt out in your settings.
  • Mind your rulesets. If the cloud agent cannot push, a branch ruleset is the usual cause.

Getting started

  1. Turn on Copilot in your editor and try agent mode on a small, well-tested change.
  2. Add a copilot-setup-steps.yml so the cloud agent can install your dependencies and run your tests.
  3. Assign one small issue to Copilot and review the pull request it opens.
  4. Write one custom agent for a job you repeat, such as tests or documentation.
  5. Keep your project rules in one place. Our free Vibe Coder Starter Kit includes starter rules files and checklists you can adapt.

Comparing Copilot with other agents? See Qodo vs Copilot, Sourcegraph Cody vs Copilot and the GitHub Copilot profile on Agentdex.

Model Pulse

Daily AI model drops, tools, and what they mean for AI coders.

FAQ

What are GitHub Copilot agents?

They are the parts of Copilot that carry out multi-step tasks rather than suggest single lines: agent mode in your IDE, the cloud agent on GitHub, the Copilot CLI and its built-in agents, and custom agents you define in Markdown profiles.

What is the difference between agent mode and the cloud agent?

Agent mode works in your local editor while you watch. The cloud agent works on GitHub in a temporary environment, runs tests and linters, and returns a branch or pull request for you to review.

Can I use Copilot agents for free?

Yes, with limits. The Free plan includes limited chat and agent use, including agent mode and the CLI, plus 2,000 completions a month. Heavier use needs a paid plan from $10/month.

How long can the Copilot cloud agent run?

Up to 59 minutes per session. You can set a shorter limit with timeout-minutes in copilot-setup-steps.yml.

Where do custom agent files go?

Repository agents are Markdown files with YAML frontmatter, usually in the .github/agents folder. Organization and enterprise agents can also be shared, and the Copilot CLI supports personal profiles.

Does the cloud agent work with GitLab or Bitbucket?

No. GitHub says the cloud agent only works with repositories hosted on GitHub.

Written by Agentdex Editorial. Agentdex is an independent directory reviewing AI coding tools; we are not affiliated with the companies mentioned. Prices and features are as of October 24, 2026, from official sources linked above. Always confirm on the vendor's site.