mirror of
https://github.com/harivansh-afk/sandbox-agent.git
synced 2026-04-15 20:03:11 +00:00
Add comprehensive tracing instrumentation across the entire agent startup path (gigacode CLI, ACP HTTP adapter, agent installation, and process spawning) to enable detailed performance profiling. Replace npm-based agent process launchers that use npx (incurring resolution overhead on every spawn) with pre-installed npm packages, reducing startup latency. Improve error diagnostics when agent processes crash by capturing exit codes and stderr tails. Update error handling to map exited processes to dedicated error variants with actionable error messages. Co-authored-by: Claude Haiku 4.5 <noreply@anthropic.com> |
||
|---|---|---|
| .. | ||
| src | ||
| Cargo.toml | ||
| README.md | ||
Supports Claude Code, Codex, Pi, and Amp.
This is not a fork (and never will be).
It's powered by Sandbox Agent SDK's wizardry.
Experimental & just for fun.
Issues — Discord — Supported OpenCode Features
How It Works
┌─ Gigacode ────────────────────────────────────────────────────────┐
│ ┌─────────────────┐ ┌─────────────────┐ ┌─────────────────┐ │
│ │ OpenCode TUI │───▶│ Sandbox Agent │───▶│ Claude Code / │ │
│ │ │ │ │ │ Codex / Pi / Amp │ │
│ └─────────────────┘ └─────────────────┘ └─────────────────┘ │
└───────────────────────────────────────────────────────────────────┘
- Sandbox Agent SDK provides a universal HTTP API for controlling Claude Code, Codex, Pi, and Amp
- Sandbox Agent SDK exposes an OpenCode-compatible endpoint so OpenCode can talk to any agent
- OpenCode connects to Sandbox Agent SDK via
attach
OpenCode Models vs Gigacode Agents
- OpenCode supports switching between inference providers (Anthropic, OpenAI, etc.). This is OpenCode talking directly to the models with its own tools, system prompts, and agentic loop.
- Gigacode automates other coding agent harnesses, so it's using the exact same logic that you would if you ran Claude Code, Codex, Pi, or Amp natively.
OpenCode (native): Model → OpenCode's tool loop → result
Gigacode: Model → Claude Code / Codex / Pi / Amp CLI → result
This means you get each agent's specialized capabilities (such as Claude Code's Read/Write/Bash tools, Codex's sandboxed execution, and Amp's permission rules) rather than a single tool loop with different models behind it.
Install
macOS / Linux / WSL (Recommended)
curl -fsSL https://releases.rivet.dev/sandbox-agent/latest/gigacode-install.sh | sh
npm i -g
npm install -g @sandbox-agent/gigacode
gigacode --help
bun add -g
bun add -g @sandbox-agent/gigacode
# Allow Bun to run postinstall scripts for native binaries.
bun pm -g trust @sandbox-agent/gigacode-linux-x64 @sandbox-agent/gigacode-linux-arm64 @sandbox-agent/gigacode-darwin-arm64 @sandbox-agent/gigacode-darwin-x64 @sandbox-agent/gigacode-win32-x64
gigacode --help
npx
npx @sandbox-agent/gigacode --help
bunx
bunx @sandbox-agent/gigacode --help
Note: Windows is unsupported. Please use WSL.
Usage
TUI
Launch the OpenCode TUI with any coding agent:
gigacode
Web UI
Use the OpenCode Web UI to control any coding agent from the browser.
OpenCode SDK
Use the @opencode-ai/sdk to programmatically control any coding agent.