A small connector for local tools串接本機工具的小型連接器

Keep the chat client. Connect it to agy.

保留原本的聊天工具,接上 agy。

Point a compatible client at a local chat-completions endpoint. The server passes requests to your authenticated Antigravity CLI and manages selection across your configured accounts.

把相容客戶端指向本機的 chat-completions 端點。伺服器把請求交給已登入的 Antigravity CLI,並在設定的帳號之間選擇。

Python 3.10+127.0.0.1:18795Source-only utility以原始碼提供
Connection example連線示意
http://127.0.0.1:18795/v1/chat/completionsCompatible client → local server → agy CLI相容客戶端 → 本機伺服器 → agy CLIUses the accounts you configure使用你設定的帳號
What it connects串接方式

Three parts, one local endpoint.

三個部分,一個本機端點。

01

Chat-completions interface

聊天介面

Accept OpenAI-style chat requests and return normal or streamed responses through the CLI bridge.

接收 OpenAI 格式的聊天請求,透過 CLI 橋接回傳一般或串流回覆。

02

Account selection

帳號選擇

Choose round-robin or fill-first selection with per-account quota cooldowns. Authenticate each account through the CLI before adding it.

選擇輪替或優先填滿策略,依帳號處理配額冷卻。加入帳號前,先透過 CLI 完成登入。

03

Local operation

本機執行

Run the standard-library Python server directly or use the supplied systemd user-service template. The default bind is loopback.

直接執行使用標準函式庫的 Python 伺服器,或使用隨附的 systemd 使用者服務範本。預設只綁定本機迴圈位址。

Start here從這裡開始

Install the CLI, then connect the client.

先準備 CLI,再串接客戶端。

You need Python 3.10+ and an installed, authenticated agy CLI. Copy the example configuration, list your account directories and start the server using the README instructions.

需要 Python 3.10 以上,以及已安裝、已登入的 agy CLI。複製範例設定、列出帳號目錄,再依 README 啟動伺服器。

Configure accounts

設定帳號

Keep each account configuration in its own private directory outside the repository.

每個帳號的設定放在儲存庫之外的獨立私人目錄。

Start and inspect

啟動並檢查

Start the server and inspect /health before connecting a chat client.

啟動伺服器,先檢查 /health,再連接聊天客戶端。

Connect locally

從本機連線

Set the client base URL to http://127.0.0.1:18795/v1 and test one request.

將客戶端 base URL 設為 http://127.0.0.1:18795/v1,先測試一個請求。

Current scope目前範圍

A focused bridge, available as source.

範圍明確的橋接工具,以原始碼提供。

An early utility with no tagged release. Core client code is vendored from soyelmismo/hermes-antigravity-subscription under MIT; this repository adds the pool, server wrapper and systemd packaging.

尚無 tag 發行版的早期工具。核心客戶端採用 soyelmismo/hermes-antigravity-subscription 的 MIT 程式碼;這個儲存庫加入帳號池、伺服器封裝與 systemd 設定。

Works today

目前可用

  • Chat-completions, model-list and health endpoints are present in the source.
  • 原始碼提供聊天完成、模型清單與健康檢查端點。
  • Both streamed and non-streamed replies, account selection and error mapping.
  • 支援串流與非串流回覆、帳號選擇及錯誤對應。

Limits and not yet

限制與尚未完成

  • This is an OpenAI-compatible subset, not the complete OpenAI API. Tool calls are emulated and CLI host tools are denied.
  • 這是 OpenAI 相容介面的子集,不是完整 OpenAI API。工具呼叫以模擬方式處理,CLI 的主機工具會被拒絕。
  • Requires your own authenticated CLI accounts and remains subject to their service access and quotas.
  • 需要自己已登入的 CLI 帳號,並受原服務的存取權限與配額限制。
  • Keep the default loopback deployment. The wrapper supplies no production authentication layer for public exposure.
  • 維持預設的本機部署方式;這層封裝沒有提供對外公開服務所需的正式驗證層。
Distribution取得方式
Source only原始碼
License授權
MIT
Runtime執行環境
Python 3.10+ · agy CLI
Default address預設位址
127.0.0.1:18795
Reviewed查核日期
2026-10-23

More from Ted Huang. Every public project has a page like this one, in English and Traditional Chinese.

Ted Huang 的其他作品。每個公開專案都有一頁像這樣的中英雙語介紹。

All projects →全部專案 →