One binary · One local proxy
Your AI configs,
everywhere.
Reuse the providers you already have — OpenCode, CrabCode, OCX, Hermes, Copilot — and wire them into Codex, GitHub Copilot App, Zed, JetBrains, Claude Code, and Grok Build. One fast native Rust binary. No second dashboard.
curl -fsSL https://joocode.yan.ad/install | bash Then run jcx — the flagship command. joocode ships as a compatible alias.
Why Joocode?
Config-first
Discovers providers you already have — OpenCode, CrabCode, OCX profiles, Hermes Agent, GitHub Copilot, and a flat OpenAI-compatible file.
Desktop-aware
Detects installed clients and prepares only the integrations present on your machine. No second dashboard to configure.
One native process
A single Rust binary, one shared proxy. No Node or Bun runtime, no browser dashboard required.
Provider control plane
Detected sources, custom providers, API-key pools, and virtual combos in one built-in terminal UI.
Protocol bridge
Responses API and Chat Completions, JSON and SSE, images, tool calls, streamed arguments, and function results.
Credential-safe
Upstream credentials stay in their original source. Joocode never prints authorization values.
How it works
Joocode builds one source-aware model registry, exposes it through a local
OpenAI-compatible endpoint at http://127.0.0.1:10100/v1, and routes
each request back to the provider that owns the model.
OpenCode ────────┐
OCX ──────────────┤ ┌─ Codex Responses API
Hermes ───────────┤ ├─ GitHub Copilot App BYOK
GitHub Copilot ───┼─► Joocode proxy ───┼─ Zed / JetBrains
Antigravity ──────┤ ├─ Claude Code Messages API
providers.json ───┘ └─ Grok Build custom models
Model IDs stay explicit: provider/model, copilot/model,
antigravity/gemini/model, or a virtual combo/name.
Desktop integrations
Probabilities estimate how likely each integration is to stay reliable across upstream client releases — not uptime guarantees.
| Target | Probability | Current behavior |
|---|---|---|
| Codex | 98% | Full catalog and Responses API integration |
| Zed | 95% | OpenAI-compatible catalog plus local credential registration |
| GitHub Copilot App | 90% | Automatic BYOK provider/model sync |
| Grok Build | 90% | Custom models written to ~/.grok/config.toml |
| Claude Code | 85% | Anthropic Messages gateway, models, JSON/SSE and tools |
| JetBrains | 80% | OpenAI-compatible endpoint via the built-in provider |
| Antigravity | 80%/70% | Official Gemini source plus an opt-in patched app |
Fast by design
| Metric | Joocode | OpenCodex / ocx | Difference |
|---|---|---|---|
| Startup time | 5.8 ms | 289.1 ms | ~49.8× faster |
| Memory, idle | 5.8 MB | 15.3 MB | ~62% less |
| Bundle footprint | 8.2 MiB | 110 MiB | ~13.4× smaller |
CLI-only runtime footprint, measured on an Apple M1 (arm64, macOS 27), September 2026. See the README for methodology.