Sub-10ms Latency • < 12MB RAM • 35 CLI Agents • A2A Protocol • Zero Node.js Overhead

Quickstart • Architecture • Supported Agents • Benchmarks • Features • Cheatsheet

GoRoute is an ultra-fast, single-binary LLM gateway engineered specifically for developers using autonomous AI coding agents (OpenCode, Claude Code, Cursor, Codex CLI, Aider, Cline, etc.).

Instead of running bloated Node/Electron proxies that consume 600MB to 1.5GB of RAM, GoRoute delivers identical multi-provider pooling, Google Antigravity OAuth rotation, Anthropic-to-OpenAI protocol normalization, and response caching inside a tiny 5.1MB footprint.

┌─────────────────────────────────────────────────────────────┐ │ YOUR CODING AGENTS │ │ OpenCode • Claude Code • Cursor • Codex CLI • Aider • Roo │ └──────────────────────────────┬──────────────────────────────┘ │ http://127.0.0.1:8128 (OpenAI /v1 & Anthropic /v1/messages) │ ▼ ┌───────────────────────────────────────────────────────────────────────────────────────────┐ │ ⚡ GOROUTE CORE │ │ │ │ ┌───────────────────────┐ ┌───────────────────────┐ ┌───────────────────────────────┐ │ │ │ Dual-Protocol SSE │ │ 55 RTK Filters │ │ Session Account Affinity │ │ │ │ OpenAI ⇆ Anthropic │ │ 70-90% Noise Cut │ │ Prompt Cache Preservation │ │ │ └───────────────────────┘ └───────────────────────┘ └───────────────────────────────┘ │ │ │ │ ┌─────────────────────────────────────────────────────────────────────────────────────┐ │ │ │ Autonomous Token Health Scheduler & Router │ │ │ │ (Cascades across accounts on 429 quota or 503 limits; proactive refresh) │ │ │ └─────────────────────────────────────────────────────────────────────────────────────┘ │ └──────────────────────────────────────────────┬────────────────────────────────────────────┘ │ ┌───────────────────────────────┼───────────────────────────────┐ ▼ ▼ ▼ ┌───────────────────────────┐ ┌───────────────────────────┐ ┌───────────────────────────┐ │ Google Antigravity Pool │ │ AWS CodeWhisperer (Kiro) │ │ OpenAI / Codex Direct │ │ Multi-Account • Sonnet │ │ Context Auto-Pruning │ │ OpenAI Compatible API │ └───────────────────────────┘ └───────────────────────────┘ └───────────────────────────┘

🚀 Quickstart

1. Install (Choose One)

📦 Option A: via npm / npx (Zero Go needed)

npm install -g goroute

Or run directly via npx:

npx goroute setup opencode

⚡ Option B: 1-Line Native Installer (Linux & macOS)

curl -fsSL https://raw.githubusercontent.com/ziuus/goroute/main/install.sh | bash

🍺 Option C: Homebrew (macOS & Linux)

brew install ziuus/goroute/goroute

🛠️ Option D: Go Toolchain

go install github.com/ziuus/goroute/cmd/goroute@latest

2. Connect Your Accounts

Sign in with Google Antigravity (PKCE OAuth):

goroute auth login antigravity

Securely links your Antigravity accounts into GoRoute's encrypted local storage (~/.goroute/vault.json).

Auto-configure your favorite agent with one command:

Configure OpenCode (writes config + registers provider credentials)

goroute setup opencode

Configure Claude Code

goroute setup claude

Configure OpenAI Codex CLI

goroute setup codex

Configure ALL 18 supported agents in one click

goroute setup all

4. Run Seamlessly

Run GoRoute as a background service:

Enable Linux background autostart (systemd user unit, ~5MB RAM):

goroute autostart enable

Verify gateway status:

goroute status

Or launch any agent with zero configuration using the universal proxy injector:

goroute run opencode goroute run claude goroute run aider --model openai/claude-3-7-sonnet

📊 Benchmarks vs Traditional Gateways

Tested on Linux x86_64 under 500 concurrent streaming requests:

goroute setup automatically configures paths, headers, authentication keys, and provider manifests for:

Universal Multi-Protocol Gateway:

POST /v1/chat/completions (OpenAI format + SSE chunks)

POST /v1/messages (Native Anthropic format + content_block_delta SSE streaming)

POST /api/chat (Native Ollama emulation)

GET /mcp (Model Context Protocol native SSE server)

GET /.well-known/agent.json & POST /api/a2a/tasks (A2A Protocol v0.3)

Zero-Alloc In-Memory Cache:

Repeated agent queries (e.g. title generation, linters, duplicate prompts) hit the internal cache and return X-Cache: HIT with 0ms latency and 0 upstream tokens.

DuckDuckGo Web Search & Readability Scraper:

Built-in POST /v1/web/fetch and GET/POST /v1/search with zero API keys required. Converts fetched HTML to clean GitHub-flavored Markdown.

Smart Quota Cascade & Multi-Account Pools:

Dynamically rotates across accounts with configurable strategies: Priority / Failover, Round-Robin, and Fair-Share.

Automatically isolates accounts on HTTP 429 quota exhaustion or 402 billing limits with proactive token cooldowns.

Context Pruning & Token Optimization:

Single-pass $O(N)$ integer pruning algorithm trims duplicate prompt prefixes and keeps multi-turn transcripts within upstream limits without dropping active agent instructions.

🤝 Contributing & Good First Issues

We welcome contributors! Check out our open Good First Issues:

GoRoute provides pre-compiled, optimized static binaries for all major platforms:

Gateway Management

goroute serve # Start gateway server (port 8128) goroute serve --no-dashboard # Headless mode (< 10MB RAM) goroute autostart enable # Enable system background service goroute autostart status # Check service status goroute autostart disable # Stop & disable service goroute status # Health check & model listing

Agent Setups

goroute setup all # Configure all 14 coding agents goroute setup <agent> # Configure a specific agent (e.g. opencode, claude) goroute run <agent> [args...] # Launch agent with proxy env variables

Providers & Accounts

goroute auth login antigravity # Interactive PKCE OAuth login goroute auth list # List connected provider accounts goroute catalog # View 25+ built-in provider presets

📄 License

MIT © ziuus. Designed for developers who love speed.