Renn Copilot

Injects Gemini, Anthropic (Claude), and GPT models into GitHub Copilot Chat in VS Code, via CLIProxyAPI and its OAuth logins (Antigravity, Claude web/Claude Code, Codex, plus Gemini CLI/Qwen/iFlow with a caveat below).

Everything lives inside a single, self-contained VS Code extension — no separate terminal, no browser tab, no other process to start by hand:

The extension is intentionally the only thing that touches chatLanguageModels.json, and only through reading/writing that one file directly, replacing just the single entry it owns — never clobbering anything else you've configured there yourself.

This is a personal, local-first tool — the backend runs on your own machine (spawned by the extension), not as a public multi-tenant service.

Dashboard

• Command Palette → "Renn Copilot: Open Dashboard" — the full dashboard as an editor tab:

  Page	What it does
  Overview	Install/update the CLIProxyAPI binary, start/stop/restart it, at-a-glance status, setup checklist, health monitor, token trend.
  Providers & Login	OAuth login per provider, stored-credentials list with per-credential and bulk enable/disable, quota reset.
  Models	Toggle which models are exposed to Copilot Chat, per-provider and global enable/disable, search/filter, vision-capability verification.
  Usage	Token usage by provider/model (sortable, filterable, with cost estimate), account health, OAuth/API key usage.
  Logs	Live tail of CLIProxyAPI's own request log and the backend's own process log, with search, copy, and download.
  Config	Raw config.yaml editor (hidden by default), routing strategy (round-robin / fill-first), discard/save.

• Activity Bar icon — a compact sidebar (deliberately not a squeezed-down copy of all 6 pages): server status with Start/Stop/Restart, a one-line health summary, enabled model count linking straight to the Models page, and quick buttons for Sync Models / Copy API Key / Open Full Dashboard.

Setup

Requires Node.js 18+ only if you're building from source. End users just install the .vsix (Extensions panel → "..." → Install from VSIX) — the backend's dependencies ship inside it, so nothing else needs installing on the machine running VS Code.

Building from source:

npm install
npm install --prefix backend
npm install --prefix webview-ui

Then, from the repo root:

npm run package   # builds the webview, installs backend deps, produces a .vsix via vsce

Install the generated .vsix (Extensions panel → "..." → Install from VSIX), or run it from source via the VS Code Extension Development Host (press F5 with this repo open as the workspace) while iterating.

Settings

Commands

• Renn Copilot: Open Dashboard — opens the dashboard as an editor tab (also available via the Activity Bar sidebar icon).
• Renn Copilot: Start Backend / Stop Backend — manual control, mainly useful when autoStartBackend is off.
• Renn Copilot: Sync Models from Dashboard — re-syncs the enabled model list into Copilot's BYOK setting.
• Renn Copilot: Copy API Key to Clipboard — for pasting into VS Code's "Chat: Manage Language Models" dialog.
• Renn Copilot: Show Provider Account Health — quick-pick breakdown of every stored credential's status.

Typical flow

1. Install the extension. The backend (and, by default, the CLIProxyAPI server once its binary is installed) spawn automatically.
2. Open the dashboard (Command Palette or Activity Bar icon). On first run, click "Install / Update binary" once on the Overview page.
3. On Providers & Login, click "Login" for each provider you use — this opens the OAuth page in your browser; the dashboard polls until the token lands.
4. On Models, toggle which models should be exposed to Copilot Chat.
5. Reload VS Code (models sync on startup by default, or run "Renn Copilot: Sync Models" manually) — whenever the synced model list actually changes, the API key is copied to your clipboard automatically. Open Copilot Chat's model picker → "Manage Models..." and click the eye icon next to the new entries to enable them; when VS Code prompts for the API key, just paste (Ctrl+V / Cmd+V) and press Enter.

Troubleshooting

• Logs page shows nothing for the "CLIProxyAPI" tab, with Management API GET /logs failed: 400 logging to file disabled in the backend's console. Harmless — CLIProxyAPI itself has request-logging-to-file turned off in its config.yaml. Either ignore it (the Backend tab on the same page still shows our own process log), or enable it via the dashboard's Config page by adding a logging-to-file: true key and saving.
• Claude Opus 4.7+ requests fail with temperature is deprecated for this model. A known, still-unpatched upstream CLIProxyAPI issue as of mid-2026 — it doesn't strip deprecated sampling params (temperature/top_p/top_k) per-model before forwarding. logging-proxy.js at the repo root is a small standalone diagnostic/hotfix proxy that sits between VS Code and CLIProxyAPI, strips those params only for the affected models, and logs the raw request/response for debugging. Run with node logging-proxy.js, then point the affected model entries in Copilot's BYOK settings at port 8318 instead of 8317 temporarily, and revert once upstream fixes it.

Known gaps

• Gemini CLI / Qwen / iFlow OAuth: CLIProxyAPI's Management API only exposes ready-made OAuth-URL endpoints for antigravity, anthropic (Claude), and codex (see backend/src/management-client.js). The other three providers currently require driving CLIProxyAPI's own CLI --login flags directly on the machine running the backend; that flow isn't wired into the dashboard yet. backend/src/routes.js returns a clear 400 explaining this if you try to log in to one of them from the UI.
• No automated tests yet — changes are currently verified by tsc (extension host + webview) and manual smoke-testing against a running CLIProxyAPI instance.