damiao-proxy · 大妙 AI Local Proxy

English (default) · 简体中文

Turn Tencent CloudBase / Hunyuan models into a local OpenAI-compatible API at http://localhost:8765/v1, with a desktop control panel and multi-account billing routing. No WeChat DevTools required — one process, one port, works with any OpenAI-format client.

---

What it does

Hosted AI models behind Tencent's 小程序成长计划 (Growth Plan) token package can only be called through the CloudBase SDK — a plain HTTP call fails with AI_CHANNEL_NOT_ALLOWED. This project solves that by relaying through a CloudBase cloud function, then re-exposing everything as a standard OpenAI endpoint on localhost, so any agent can consume it.

Your agent (OpenAI format)
    │  POST /v1/chat/completions
    ▼
damiao-proxy  ── http://localhost:8765/v1 ──┐
    │                                       │
    ├── ① CloudBase SDK (anonymous login) ──┤──▶ Hunyuan / Qwen models
    └── ② Cloud function `ai-proxy` (HTTP) ─┘    (text · image · I2I)

• Channel ① — SDK direct call: supports streaming, function calling (FC) and reasoning_content. Preferred.
• Channel ② — Cloud function fallback: non-streaming, compatible with older deployments.

Features

• OpenAI-compatible — /v1/chat/completions (stream + non-stream), /v1/models, /v1/images/generations.
• Desktop app — Electron shell with system tray, portable .exe build, port auto-release on exit.
• React control panel — live model catalog, connectivity test, account manager, metrics & charts, image workspace, conversation history.
• Multi-account routing — Growth-Plan token account and resource-point account side by side; per-request account selection.
• Function calling — tools / tool_choice passthrough with an agent tool loop (maxIter).
• Vision & resource bridge — image input routed to a vision-capable model; PDF / text files extracted and injected into the prompt.
• Image generation — text-to-image and image-to-image, with history and local caching.
• Encrypted local secrets — apiKey values are stored AES-256-GCM encrypted, derived from the machine fingerprint.

Quick start

Run from source

git clone https://github.com/deity6/damiao-proxy.git
cd damiao-proxy
npm install
cp config.example.json config.json     # then fill in your CloudBase envId
node server.js

Then open http://localhost:8765/ for the control panel. On Windows you can also just double-click 启动代理.bat.

Desktop app (Electron)

npm install                 # installs electron + electron-builder (devDependencies)
npm start                   # launch the desktop shell (--debug opens devtools)
npm run dist:portable       # build a portable Windows .exe into release/

Connect your agent

Field	Value
Base URL	http://localhost:8765/v1
API Key	any string (e.g. free-ai) — not validated
Default model	hy3

Works with anything that speaks the OpenAI protocol: codex, claude-code-switch, WorkBuddy custom models, openai SDK clients, and so on.

curl

curl http://localhost:8765/v1/chat/completions \
  -H "Content-Type: application/json" \
  -H "Authorization: Bearer free-ai" \
  -d '{"model":"hy3","messages":[{"role":"user","content":"Say hi in one sentence"}]}'

Python

from openai import OpenAI

client = OpenAI(base_url="http://localhost:8765/v1", api_key="free-ai")
r = client.chat.completions.create(
    model="hy3",
    messages=[{"role": "user", "content": "Say hi in one sentence"}],
)
print(r.choices[0].message.content)

Streaming

curl -N http://localhost:8765/v1/chat/completions \
  -H "Content-Type: application/json" \
  -d '{"model":"hy3","stream":true,"messages":[{"role":"user","content":"Count to five"}]}'

Model catalog

GET /v1/models returns the live list. Text and image generation share the same token package.

Growth-Plan token account (default, cloud-function channel)

Type	Model id	Group	Notes
Text	hy3	hunyuan-exp	Hunyuan Hy3 (295B/21B MoE, 256K ctx). 3-level deep thinking + FC. Default
Text	hunyuan-free	hunyuan-exp	Friendly alias of hy3, same underlying model
Text	hy3-preview	hunyuan-exp	Preview entry, auto-routed to hy3
Image	HY-Image-3.0-Plus-4090-Tob-v1.0	hunyuan-image	Text-to-image (default), supports aspect ratio / revise / thinking
Image	HY-Image-v3.0-I2I-ToB-v1.0.1	hunyuan-image	Image-to-image (reference image + prompt)
Image	hunyuan-image, hunyuan-image-v3.0-v1.0.4	hunyuan-image	Retired upstream on 2026-07-15 — migrate to the v3.0 Plus model

Resource-point account (gateway channel)

Type	Model id	Vision	Thinking	FC
Text	hy3, hy3-preview	–	✅	✅
Text	qwen3.5-flash, qwen3.5-plus	✅	flash ✗ / plus ✓	✅
Text	hunyuan-role-latest, hy-role	–	✗	✅
Text	hy-mt2-lite, hy-mt2-plus, hy-mt2-pro	–	✗	✗
Image	HY-Image-3.0-Plus-4090-Tob-v1.0 (T2I), HY-Image-v3.0-I2I-ToB-v1.0.1 (I2I)	–	–	–

Image generation is always served through the cloud function (hunyuan-image group) — the gateway cloudbase group returns 403 AI_MODEL_NOT_SUPPORTED for images.

API surface

OpenAI-compatible

Method	Path	Description
POST	/v1/chat/completions	Chat completion, stream supported. Aliases: /chat/completions, /v1
GET	/v1/models	Live model catalog for the active account set
POST	/v1/images/generations	Text-to-image and image-to-image

Control panel (used by the React UI)

Method	Path	Description
GET	/api/status	Service state, port, active account, environment info
GET PUT	/api/settings	Read / update settings (port, defaults, bridges)
POST	/api/toggle	Start or stop the proxy
POST	/api/test	Connectivity test (text or image)
GET POST	/api/accounts	List / create accounts
PUT DELETE	/api/accounts/:id	Update / delete an account
GET	/api/accounts/trash	Deleted accounts awaiting restore
POST	/api/accounts/:id/restore	Restore a deleted account
POST	/api/accounts/test · /api/accounts/probe	Test or probe an account's models
GET PUT	/api/changelog	Changelog data
GET	/api/metrics	Request / success / failure / token metrics
GET POST	/api/conversations	Conversation list / create
GET PUT DELETE	/api/conversations/:id	Read / update / delete a conversation
POST	/api/system/open-path · /api/system/pick-directory · /api/system/migrate-storage	Filesystem helpers
GET	/api/logs	Recent server logs
GET	/api/images/history · /api/images/:name	Generated image history and files

Configuration

Copy config.example.json to config.json and fill in your own values. config.json is git-ignored — never commit it.

Key	Meaning
envId	CloudBase environment ID
functionName	Cloud function name (default ai-proxy)
deviceId	Device identifier used for anonymous login
port	Local port (default 8765; auto-increments if occupied)
defaultModel / modelAlias	Real default model / the alias exposed to clients
proxyEnabled	Whether the proxy starts automatically
visionBridge	Route image inputs to a vision-capable model (eyeModel)
resourceBridge	Extract PDF / text attachments (maxFileBytes, eyeModel)
agentTools	Agent tool loop: enabled, unsafeTools, maxIter
accounts	Account list (type: token | resource)
routing.defaultAccount	Default account id

Deploy the cloud function

The relay cloud function lives in cloudfunctions/ai-proxy.

cd cloudfunctions/ai-proxy
npm install
# set your own envId in cloudbaserc.json, then deploy with tcb / CloudBase CLI

Requirements: Nodejs 18.15 runtime, anonymous login enabled, security rule allowing anonymous invocation of ai-proxy only.

Development

# Control panel (Vite + React)
cd ui
npm install
npm run dev       # dev server
npm run build     # build to ../dist, served automatically by server.js

# Portable desktop build
npm run dist:portable

Project structure

damiao-proxy/
├── server.js            # HTTP server: OpenAI endpoints + control-panel API + account routing
├── electron-main.js     # Electron shell, tray icon, port lifecycle
├── preload.js           # Renderer bridge
├── models.js            # Data-driven model catalog (token / resource-point)
├── config.example.json  # Config template (no secrets)
├── changelog.json       # Version history (source of CHANGELOG.md)
├── 启动代理.bat          # Windows one-click launcher
├── launcher.ps1         # PowerShell launcher
├── assets/              # Icon, tray icon, splash
├── ui/                  # React control panel (Vite)
├── mcp/                 # Optional MCP server for image generation
└── cloudfunctions/
    └── ai-proxy/        # CloudBase relay function (text + image)

Privacy & security

• Everything runs locally. No telemetry, no third-party analytics.
• Account apiKey values are encrypted with AES-256-GCM using a machine-derived key; copied files cannot be decrypted elsewhere.
• config.json, data/, rp-key.txt and all logs are git-ignored by default.
• Model prompts and responses go to Tencent CloudBase — follow their terms when using Growth-Plan tokens.

Known limitations

• hy3 relayed through the proxy may return an empty content with only reasoning_content populated — an upstream quirk, not a proxy bug.
• hunyuan-2.0-instruct does not support function calling natively; use hy3 if you need tools.
• Image generation returns 422 when the upstream content filter rejects a prompt — rephrase and retry.
• The gateway (cloudbase) channel does not serve image models; images always go through the cloud function.

Changelog

See CHANGELOG.md.

License

MIT — see LICENSE.

This is an unofficial personal project. It is not affiliated with Tencent CloudBase or Hunyuan. Use it in compliance with the platform's terms of service.