Licoy/GoAmzAI: a private AIGC platform you host yourself

What problem GoAmzAI claims to solve, and for whom

GoAmzAI describes itself as a private AIGC platform for individuals, teams and enterprises. The problem it targets is not model access. It is operations. Anyone can call an OpenAI-compatible endpoint. Running a service where many users log in, spend points, generate images, submit documents for parsing and pay for top-ups is a different job, and that is the job this project packages.

The README lists the commercial machinery explicitly: a points system split across conversation, drawing, video, music, PPT, PDF and general categories; a package system with daily resets or fixed-period totals; redemption codes; an invitation mechanism with detection for users inviting themselves; a referral commission module; payment through WeChat, Alipay, Hupijiao, Rainbow EasyPay and custom external links. That list tells you who this is for. It is for someone running a paid or internal AIGC portal, not someone who wants a chat window.

Three editions are named: Plus, Pro and Enterprise. The comparison table marks custom home page, multi-device login limits, differentiated points types, marketing notifications, log cleanup, online sharing and a canvas template marketplace as unavailable in Plus, while a private knowledge base with its own ELT cleaning channel and vector database is Enterprise only. The README also states that the table excludes discontinued editions and is not the complete feature list.

Architecture: Gin on the back, Vue3 on the front, MySQL and Redis underneath

The stack is stated plainly. Front end is Vue3 with NaiveUI. Back end is Golang with Gin. Data support is MySQL 5.7 and Redis. There is no queue broker, no object storage service and no vector database in the required list, which is consistent with the claim that the minimum server is 1 core and 1 GB of memory and that basic usage stays under 100 MB of resident memory. Those are the project's own figures, not measurements taken here.

The operational design leans on channel pools. The README describes a conversation account pool covering OpenAI, 文心一言, 讯飞星火, 智谱, DeepSeek, OpenAISB, API2D and OneAPI, and separate channel management for video, music and PPT where each API key and channel record is managed independently. Drawing channels get their own concurrency thread setting with thread isolation, plus online start, restart and shutdown with a live queue count. That is a scheduling layer sitting between your users and upstream providers.

Content safety is layered rather than single-pass: a built-in word list, then a custom word list, then optional third-party moderation from Baidu Cloud or Tencent Cloud, with the layers able to run at the same time. Plugins are the other extension point. The README says 18 or more are available, covering web search, IP lookup, weather and courier tracking, and that you can write your own in ECMAScript 5.1, PHP, Python or NodeJS. MCP tool calls are supported over SSE, StreamableHTTP and Stdio, managed from the admin panel and toggled per conversation.

Installing GoAmzAI: what the repository does and does not tell you

This is the part where you need to read carefully. The repository contains README.md and an images directory. Nothing else. There is no installer script, no Dockerfile, no docker compose file and no environment template in the tree. The README states that it holds only a basic introduction, that the content may not be current, and that the official site and a Feishu document are authoritative. So the honest answer to "how do I install it" is: not from this repository.

The README does describe the deployment shape. It says deployment is done through 宝塔 (BaoTa panel) in about five minutes, and that the configuration file needs only the database settings to start. It also lists what you must have ready before you begin: a domain, a server of at least 1 core and 1 GB of memory, optionally a conversation model API key, optionally a drawing platform subscription, and optionally accounts for SMS, email and content moderation.

What that means in practice is that the commands you will run are the ones published on d.goamzai.com or in the Feishu document, not anything in this repo. The only concrete configuration fact the README gives is that the database section is the minimum required to boot. If you are evaluating this project, treat the missing deployment files as an evaluation result in itself: you cannot read the install path before you buy or register.

The demo sites are the practical first step. The README points to a Plus demo and a Pro demo and says new accounts receive trial points, so you can exercise the admin surface before committing to a server.

Where GoAmzAI is the wrong tool

The first limitation is licensing. The repository does not state a licence, and the README does not discuss one. For a platform that includes payment handling, referral commissions and user data, an unstated licence is a real blocker for legal review. Nothing here tells you whether you may modify and redistribute it, or what happens to your obligations if you embed it in a commercial product. That is not a legal opinion, it is an observation that the information is absent.

The second is the repository-versus-product split. The README explicitly warns that the listed features may not be current and directs readers to the official site. A repository that is a brochure rather than a source of truth makes version pinning hard. You cannot diff what you are running against what the README promised.

The third is the edition gating. A private knowledge base with self-built ELT cleaning and vector database retrieval is Enterprise only, and the README notes the comparison table excludes discontinued editions. If your requirement is retrieval over your own documents, the tier that supports it is the top one, and the README does not publish pricing.

Finally, this is a large surface. Payments, SMS, email, moderation, third-party logins, MCP tools and a plugin runtime all in one deployment means a lot of moving parts on a 1 GB server. The README's 100 MB figure describes basic usage, not a fully loaded instance with moderation and plugins active.

How it compares with a plain OpenAI-compatible gateway

The closest thing to a real alternative here is a thin gateway such as OneAPI, which the README itself lists as a supported conversation account pool provider. The difference in approach is sharp. A gateway does one job: it accepts OpenAI-shaped requests and routes them to whichever upstream key or provider you configure. It has no user accounts, no points, no payment flow and no image or video panels.

GoAmzAI takes the opposite route. It assumes the gateway problem is already solved and builds the tenant layer on top: users, sessions, points, packages, redemption codes, referrals, moderation and admin panels. The account pool feature means you can point it at a gateway rather than at each provider directly, so the two are not mutually exclusive.

That framing should drive your decision. If your users are engineers who will bring their own keys, a gateway is less software to run. If your users are customers who need a login, a balance and a checkout page, GoAmzAI is addressing a problem the gateway does not attempt. The trade is operational weight: MySQL 5.7 and Redis become your responsibility, and so does every module you switch on.

Maintenance, upgrade cost and licence exposure

The repository was last pushed on 2026-07-14, which is two months before the date of this review. It is not archived. That is recent enough that the project is not dormant, but the README's own warning that its contents may lag the official site means the repository push date is a weak signal for how current the product is. Documentation, pricing and deployment instructions live off-repository.

Upgrade cost is the harder question, and the README does not answer it. There is no changelog, no release list, no migration guide and no statement about how database schema changes are delivered. The README does not document rollback. For a platform holding user balances, referral earnings and payment records, the absence of a documented migration path is the single most important thing to resolve before you put real money through it.

The licence is unstated, as noted above. That affects redistribution and, depending on the eventual terms, hosting obligations. Ask the vendor directly rather than inferring from the repository, because the repository gives you nothing to infer from.

On the operational side, the README's minimum of 1 core and 1 GB is a floor, not a target. The moment you enable third-party content moderation, MCP tools, the plugin runtime and drawing channels with their own concurrency settings, you are running a multi-service workload. Size for that, not for the basic figure.

Editorial conclusion

GoAmzAI is aimed at operators who want a branded, multi-user AIGC portal with points, payments and channel management, and who are willing to follow the official site rather than this repository. It is a poor fit if you only need a single-user chat client, if you cannot run MySQL 5.7 and Redis, or if you need an open licence and complete source before committing. Verify the licence and the version comparison for the tier you intend to buy, and confirm that the deployment path documented on d.goamzai.com matches your server before you start.

Frequently asked questions

Official sources

1. Issues
2. Licoy/GoAmzAI on GitHub
3. Project website
4. README