← 插件

ZCode

Use your installed ZCode agent in Paseo and view GLM Coding Plan usage.

$ paseo plugin add lianxin255/zcode
报告问题
lianxin255Provider4次安装d97fe7f711e9源代码
ZCode 截图 1

该插件的概览尚未翻译为你的语言,因此以英文显示。

Paseo provider plugin that adds ZCode (Z.ai's GLM coding agent) as an agent backend by wrapping the community zcode-acp-server ACP bridge. Models, modes, permissions, resume and streaming run over the standard Agent Client Protocol.

What you get

  • A zcode provider in Paseo: streaming turns, editing modes (Plan / Build / Edit / Guarded / Yolo), permission prompts, session resume, model and thinking catalogs discovered live from your ZCode install.
  • A GLM Coding Plan usage card in the Usage screen and paseo usage.
  • ZCode owns everything else: sign-in, credentials, tools, model access and conversation storage stay in ~/.zcode, exactly as in the ZCode desktop app. No API keys are configured in Paseo.

Requirements

Paseo ≥ 0.11.0 on the daemon; Node.js ≥ 22 on the daemon host; ZCode (desktop app or CLI ≥ 0.14.8) installed and signed in; plugins enabled ("pluginsEnabled": true in ~/.paseo/config.json).

This plugin registers the provider id zcode. The registry's supermomonga/zcode-provider registers the same id; install one or the other, not both.

How it works and where data goes

  1. Installation prepares the checkout with npm ci --omit=dev (committed lockfile), installing the pinned zcode-acp-server (Apache-2.0) from npm.
  2. On load, the plugin finds the bridge (its own checkout copy first, then well-known npm global roots), picks a Node ≥ 22 by probing node:sqlite, and resolves the ZCode CLI entry (zcode on PATH or the desktop-app bundle).
  3. It writes a 10-line launcher shim under ~/.paseo/plugin-launch/ (0700) that sets ZCODE_BIN/ZCODE_NODE and imports the bridge. The bridge then spawns zcode app-server --stdio; Paseo talks ACP to the bridge.
  4. Model requests and authentication are handled by ZCode itself against the Z.ai/BigModel endpoints you are signed into — the plugin adds no network hops. The only request the plugin itself makes is a GET of the official GLM quota endpoint (open.bigmodel.cn / api.z.ai) to populate the usage card. That request is sent only when the active ZCode provider's base URL is exactly one of those two official hosts; for any other provider (custom gateways, self-hosted or local endpoints) the card reports unavailable without contacting the network, so a third-party key is never attached to an official endpoint. Nothing is sent anywhere else; no telemetry.

Unofficial integration; not affiliated with or endorsed by Zhipu AI / Z.ai.

More

README.md (English) and README_CN.md (Chinese) cover configuration overrides (agents.providers.zcode.env.ZCODE_BIN), troubleshooting, platform notes and the development workflow (npm run smoke drives a real ACP session).