GitHub Copilot is the AI assistant built by GitHub, a Microsoft company, for software development. As an AI harness, it supplies the tools, context management and execution environment that turn a language model into a working coding agent, through Copilot CLI, the coding agent, and agent mode in IDEs. As of 20 September 2026.
What is GitHub Copilot?
GitHub Copilot is built by GitHub and is proprietary. Copilot CLI — described in GitHub’s documentation as bringing “the power of GitHub Copilot directly to your terminal” — entered public preview in September 2025 and has been generally available since February 2026. It runs on macOS, Linux and Windows (including PowerShell and WSL), and is installed via npm install -g @github/copilot, Homebrew, WinGet, or an install script. It requires an active Copilot subscription; an interactive session starts with the copilot command. Alongside the CLI, Copilot exists as an agentic mode in VS Code and other IDEs, and as a coding agent that works on tasks directly on GitHub.com.
GitHub Copilot as an AI harness
Copilot CLI calls the model, evaluates its tool requests, and executes them — either interactively or non-interactively via -p/--prompt for scripts and CI pipelines. A plan mode, toggled with Shift+Tab, builds a structured plan before making changes. Tool calls trigger approval prompts, answerable once, for the whole session, or with rejection; the flags --allow-all-tools, --allow-tool and --deny-tool let you pre-set these rules. Per the documentation, Copilot reads project instructions from AGENTS.md files (multiple allowed, nearest one taking precedence) as well as from .github/copilot-instructions.md. Tool integrations rely on the Model Context Protocol (MCP); custom agents allow specialised Copilot variants for individual tasks, and plugins bundle agents, skills, hooks and integrations as installable packages. Hooks can, per the documentation, log or block prompts and tool calls. For execution environments, the CLI offers local sandboxing (/sandbox enable) as well as cloud-backed sessions (copilot --cloud).
Models and model access
Copilot CLI lets you switch models with the /model slash command; available options include models from the OpenAI GPT, Anthropic Claude and Google Gemini families, depending on the Copilot plan. GitHub also documents its own BYOK (bring your own key) configuration: the environment variables COPILOT_PROVIDER_BASE_URL, COPILOT_PROVIDER_TYPE, COPILOT_PROVIDER_API_KEY and COPILOT_MODEL let you point the CLI at any endpoint compatible with the OpenAI Chat Completions API, at Azure OpenAI, at the Anthropic API, or at locally hosted models (for example via Ollama) — provided the model supports tool calling and streaming. This means GitHub Copilot CLI can be operated through a central, OpenAI-compatible proxy such as the AI Gateway by setting its endpoint as COPILOT_PROVIDER_BASE_URL.
Enterprise use: data protection and governance
GitHub offers a Data Processing Agreement for Copilot Business and Enterprise. Per the documentation, Enterprise environments can use “GitHub Enterprise Cloud with data residency” to restrict Copilot to an EU data region, keeping inference and associated data there; admins can additionally restrict models to FedRAMP-compliant options for US data residency. Organisations with an existing SAML SSO setup (Entra ID, Okta, OneLogin, PingOne) use the same sign-in flow for Copilot. Content exclusion lets you remove specific repositories or paths from Copilot’s reach, and audit logs record, per the documentation, seat assignments, policy changes and access events over 180 days. Regardless of the model configured, source code and prompts go to whichever model provider is set, which should be reflected in your own record of processing activities beforehand.
Orchestrating GitHub Copilot with herdr
Per herdr’s agent overview, GitHub Copilot CLI is listed there with the same, lighter integration level as many terminal agents without installable lifecycle hooks: herdr detects a running session via a screen manifest and attaches to it. Teams orchestrating several coding agents in parallel terminal sessions can include Copilot CLI alongside other harnesses this way.
Where it fits: developer harness and enterprise harness
GitHub Copilot is a harness for developers working in the terminal, in an IDE, and on GitHub.com. For business users outside development, who work with company knowledge rather than code repositories, CompanyGPT is the matching harness — with tools via MCP servers and integrations, context via companyRAG, and roles from your existing directory service. The AI Gateway provides the shared cost and governance layer that connects both coding harnesses such as Copilot CLI and CompanyGPT to central budgets, providers and logging. An overview of further AI harnesses is available at /en/ai-harness/. We help organisations embed GitHub Copilot into existing AI governance and a central AI Gateway.