A local bridge between AI coding CLIs and OpenAI-compatible providers.
Stop editing environment variables, configuration files, and model settings every time you switch AI coding CLIs.
Select a CLI. Select a provider. Start coding.
It automatically discovers available models and prepares the correct CLI configuration before launch.
| Feature | What it does |
|---|---|
| One provider workflow | Select a configured provider before starting a supported coding CLI. |
| Automatic model discovery & caching | Retrieve and cache models from the provider API, reuse the previous model when available, and otherwise use the first discovered model. |
| CLI-specific setup | Apply the selected provider's runtime environment and generate the required CLI configuration. |
| API-key safety | Read keys from the environment or prompt; never store them in providers.json or generated OpenCode configuration. |
| Copilot model switching | Add optional /model_byok switching through the BYOK Bridge Copilot extension. |
Every AI coding CLI expects its own environment variables, model names, and configuration format.
BYOK Bridge removes that friction by providing one provider workflow across supported CLIs.
BYOK Bridge includes an optional extension developed specifically for GitHub Copilot CLI. With the extension installed, start Copilot through the Hub, then run /model_byok to switch among the models preloaded for the selected provider. /model_byok is not a built-in Copilot CLI command. Model names and availability vary by provider.
Before installing, make sure at least one supported CLI (copilot or opencode) is available in PATH, and Node.js 22 or later is installed (required on Windows, Linux, and WSL2).
-
Install from a repository checkout:
Windows (CMD)
bin\win\install.cmd
Linux / WSL2 (Bash)
bash bin/linux/install.sh exec bash -
Configure an OpenAI-compatible provider in
<data-dir>/providers.json. Start with the example configuration or follow the Provider Configuration guide. -
Launch BYOK Bridge:
byok
For the complete first-run walkthrough, see the Quick Start guide.
- Windows
- WSL2
- Linux
Tested environments: Windows, WSL2 (Ubuntu), and Ubuntu Linux.
| CLI | Status |
|---|---|
| GitHub Copilot CLI | ✅ Experimental |
| OpenCode CLI | ✅ Experimental |
BYOK Bridge uses a generic OpenAI-compatible provider configuration rather than vendor-specific adapters. It is designed for providers that expose a model-list API: by default GET /models with model IDs at data[].id. modelsApi.path, itemsPath, and idPath can be configured when an endpoint differs.
LM Studio is the currently verified provider configuration. The providers below are common OpenAI-compatible targets; test each endpoint and model with your selected coding CLI before relying on it in a workflow.
| Provider | Typical baseUrl |
Notes |
|---|---|---|
| Ollama | http://localhost:11434/v1 |
Local OpenAI-compatible server; API key is normally not required. |
| LM Studio | http://localhost:1234/v1 |
Start its local server first; the Hub's example configuration uses this address. |
| OpenRouter | https://openrouter.ai/api/v1 |
Remote endpoint; configure an API key and select a coding-capable model. |
| vLLM | http://<host>:8000/v1 |
Self-hosted OpenAI-compatible server; endpoint and authentication depend on deployment. |
| LiteLLM Proxy | http://<host>:4000 |
OpenAI-compatible gateway; configure the proxy's key requirement and model list. |
- Quick Start guide
- Install and upgrade
- Configure providers
- Launch a CLI and use optional Copilot model switching
- Removal and development
- Changelog
- License
- API keys are read from environment variables or prompted at runtime; they are never stored in configuration or cache.
- Generated OpenCode configuration references environment variables instead of plaintext credentials.
- See the Provider Configuration guide for implementation details.