How to use Claude, Codex, and BYOK in GitHub Copilot for VS Code | GitHub Copilot Day
Summary
GitHub Copilot now supports multiple AI models (Claude, Codex, custom models) through a unified VS Code interface, with BYOK (Bring Your Own Key) enabling developers to use their own API keys, compute resources, or locally-run models—eliminating vendor lock-in while maintaining single-UI simplicity.
Key Takeaways
- BYOK (Bring Your Own Key) lets developers integrate personal API keys from OpenRouter, Anthropic, or local models directly into Copilot's agent harness, bypassing subscription requirements and supporting air-gapped networks.
- Three distinct agent harnesses (Copilot, Claude, Codex) each bring custom prompts and tool sets, functioning differently but accessible through one VS Code UI—enabling competitive evaluation without context switching.
- A single Copilot subscription grants access to all harnesses (Copilot, Claude, Codex via ChatGPT), eliminating the need for multiple vendor subscriptions while maintaining flexibility to choose preferred models.
- Developers can use free-tier alternatives like ChatGPT free subscription with Codex harness or open-source models via OpenRouter, lowering barriers to experimentation for cost-conscious builders.
- Multi-model support addresses two core freedoms: model choice (different vendors/providers) and compute ownership (local, private networks, or managed infrastructure), decoupling developer workflow from vendor constraints.
Related topics
Transcript Excerpt
Hi, I'm Tyler Leanhart. You may remember me from uh my oneliner in the VS Code documentary and also my college submission to a film festival. I'm going to be talking to you about BY and multi-agent harness in VS Code. So, what does that even mean? Uh well, it's basically two kinds of freedom, right? Copilot as you know already provides multi- uh models models from uh various different vendors openai anthropic etc. Uh but uh you don't have to get your models directly from a co-pilot subscription. We also support BY and that means bring your own key. So that means using the compute that you already have or if you're working inside of a private or air gap network uh that also works or if you run uh models locally on your machine you can al also bring those in to the co-pilot uh agent harness …
More from GitHub
- Let's Learn GitHub Copilot SDK
- How to use GitHub Copilot with WSL on Windows
- 25 agent skills to improve your workflow in GitHub Copilot | Matt Pocock | GitHub Copilot Day
- How to teach GitHub Copilot about your codebase | Tutorial for Beginners
- How to run parallel AI agents in the GitHub Copilot app | Tutorial for beginners