GeekAI: an AI creation workbench that ships the business layer too
A complete open-source AI assistant solution with a built-in admin backend, ready to use out of the box. Integrates LLMs from multiple platforms including ChatGPT, Azure, ChatGLM, iFlytek Spark and Wenxin Yiyan (ERNIE Bot). Supports plugins such as Midjourney AI art, Stable Diffusion AI art and Weibo trending topics. Built with Go + Vue3 + element-plus.
At a glance
- What is it?
- GeekAI bundles chat, image, music and video generation with credits, billing and an admin panel in one Go and Vue3 stack, started by a single Docker Compose command. The catch is that the private deployment path has no upgrade route of its own.
- Who is it for?
- GeekAI fits a team that wants a finished, sellable AI product rather than a component library, and that is ready to provision MySQL, Redis and object storage before the first request. Teams who need a documented upgrade path, per provider failure handling or a verified performance figure should read the release notes and the docs site first, because the repository itself does not promise any of those things.
- Can I use it commercially?
- Yes. Apache-2.0 is a permissive licence: you can use, modify and sell software built on it, as long as you keep its copyright and licence notices.
- Is it still maintained?
- Yes. The repository last received commits 63 days ago.
- What is it written in?
- Mainly Vue, according to GitHub's language statistics.
Answers come from the project's GitHub data, last synced on November 15, 2026, and from our analysis. They are not legal advice.
Star history
- yangjian102621/geekai4,723 ★ · +24
| Repository | 2026-09-14 | 2026-11-15 |
|---|---|---|
| yangjian102621/geekai | 4.7k | 4.7k |
24 Stars in the last 62 days, measured from the daily GitHub snapshot. Compare with other projects
Editorial analysis
docker-compose up -d is the whole install path on offer
The documented deployment is two steps: install docker and docker-compose yourself, then run one command in the project root.
docker-compose up -dThat single line starts a stack the description calls Go + Gin + MySQL + Redis on the back end and Vue3, Element Plus and Vant on the front, with Docker containerization described as one click startup and easy scaling. The repository layout backs that up: api/, build/, config/, database/, docker/, web/ and a desktop/ directory sit at the top level, alongside AGENTS.md, CLAUDE.md, PROMPT.md, CHANGELOG.md and .github/.
What the deployment section does not do is name a port, list environment variables, or show a config file you have to fill in first. There is no documented admin account bootstrap step, no seed data command, and no note about which directory holds secrets. A public demo runs at chat.geekai.me and the documentation lives at docs.geekai.me, so the README's own answer to those questions is that they are handled elsewhere.
Private deployments have no upgrade path inside the container
One line in the feature list carries more operational weight than the rest: private deployment is stated as not supporting upgrade, and requiring manual upgrade work instead. Nothing else in the deployment section explains what that means in practice.
For a self hosted operator this is the section that decides the cost. You are not describing a defect, you are describing the absence of the mechanism that would otherwise let you move from tag to tag. Any change to config/, any schema change under database/, and any new setting the newer release expects has to be worked out by hand against a running instance. The README documents no migration runner, versioned config schema or rollback step.
Tags do move. Recent releases listed for the repository are v4.3.0 on 2026-08-11, v4.2.8 on 2026-08-03 and v4.2.3 on 2025-11-26, and the last push on the default branch main is dated 2026-09-13. So the project is current while a private deployment built on an older tag stays exactly where it was put.
Credits, subscriptions and an admin panel ship in the same box
This is the part that separates GeekAI from a chat wrapper. The user system covers registration and login, permission management and credit recharge. Billing is described as supporting both pay per use and monthly subscription. Data analysis covers user behaviour, consumption records and system performance monitoring, all behind an administrator interface.
For a reader evaluating it, that changes what a trial has to answer. You are not just checking whether a prompt renders; you are deciding whether credits reconcile against upstream spend, whether a subscription locks features you cannot enumerate, and whether the statistics screen is something you would show a customer. None of those numbers appear anywhere: there is no credit price list, no exchange rate between credits and currency, and no stated rule for what happens to credits when an upstream provider returns an error.
The money path is the part most likely to need your own design, and it is also the part with the fewest documented details here.
Nine upstream engines, and no documented behaviour when one is down
Chat lists ChatGPT and Claude. Image generation lists MidJourney, DALL-E, Nano-Banana, 即梦 and 可灵, and the project description also names MJ AI painting and Stable Diffusion. Audio runs through Suno. Video lists Luma, KeLing, 即梦 and Veo3. Mind maps and a 微博热搜 plugin tool are listed as well.
Spread across those providers, the consistency question becomes yours to answer. The README does not say whether a model appears in the UI only after its key is present, whether a missing key hides the feature or shows it as broken, and whether a failed request falls back to another provider. It does not record per provider rate limits, retry policy, or whether generated images and audio are stored locally before being pushed to object storage.
Streaming responses are advertised as the reason the interface feels continuous, so the failure case matters more here than in a request and reply tool: a stream that stops halfway is a broken artefact the user already paid credits for.
Storage is delegated to a cloud bucket you have to choose yourself
File management is handled by 七牛云, 阿里云 OSS, 腾讯云 OSS or Minio. That is four separate credential sets, four billing relationships, and a real chance that your generated work ends up somewhere you did not intend.
The stated consequence for a reader is direct: if you host images in one provider and want them cached or served through a CDN attached to another, nothing in the README explains how that is configured. There is no documented bucket naming rule, no guidance on which key format each of the four expects, no visibility setting guidance, and no signed URL expiry to plan around.
Minio is the one option that keeps files inside your own network, and it is also the one you have to run and secure yourself, next to the MySQL and Redis services the compose file expects. The README gives no sign of a self hosted default, so treat it as work you own rather than a checkbox.
The 80 percent cache claim has nothing behind it
The README claims a multi layer cache strategy that improves response speed by 80 percent. It gives no baseline, no workload, no cache hit ratio, and no comparison against an uncached build, so the figure cannot be checked or reproduced from what is written here.
That does not make caching absent. It makes it undocumented. What is missing is the part you need to operate: which layers exist, what invalidates an entry, whether a cached response survives a change to a system prompt, and what a warm cache means for your own billing when the upstream price changes. A cache in front of several metered providers is also a place where a stale answer can outlive the model version that produced it.
If latency is your concern, treat the percentage as a claim about the authors' environment, and measure your own. The number gives you no basis for a capacity plan either way.
A live demo and a docs site carry the setup detail the README drops
The README points at chat.geekai.me as a running demo and docs.geekai.me for documentation, and sends feature screenshots to docs.geekai.me/info/. That division has a practical consequence for anyone evaluating this project: the parts that decide your timeline live off the repository.
The repository is not silent about its size, either. It carries 4709 stars, 1068 forks and 32 open issues under an Apache-2.0 license, with Vue as the primary language. Feature lists and screenshots tell you what the interface looks like; they do not tell you how a provider key is scoped, what a credit costs, or how to reverse a partial upgrade.
For a team that intends to run this privately, that gap is worth closing before you commit. Ask the documentation for the config surface, the billing rules and the upgrade procedure, and confirm your answers against a running instance, because the release tags alone will not tell you whether a private deployment moved forward.
Editorial conclusion
GeekAI fits a team that wants a finished, sellable AI product rather than a component library, and that is ready to provision MySQL, Redis and object storage before the first request. Teams who need a documented upgrade path, per provider failure handling or a verified performance figure should read the release notes and the docs site first, because the repository itself does not promise any of those things.
Frequently asked questions
what is geek ai
GeekAI is an open source AI creation platform written in Go with a Vue3 front end, under an Apache-2.0 license. It bundles chat, image, music and video generation with credits, subscriptions and an admin panel, and its documentation describes private deployment with no built in upgrade path.
How do I start GeekAI locally?
Install docker and docker-compose, then run docker-compose up -d in the project root. The stack is described as Go, Gin, MySQL and Redis with a Vue3, Element Plus and Vant front end, deployed as Docker containers.
Which AI engines does GeekAI connect to?
Chat covers ChatGPT and Claude, image generation covers MidJourney, DALL-E, Nano-Banana, 即梦 and 可灵, audio runs through Suno, and video covers Luma, KeLing, 即梦 and Veo3. Mind maps and a 微博热搜 plugin are listed as well.
Does a private GeekAI deployment support upgrades?
No. The feature list states that private deployment does not support upgrade and that upgrading has to be done by hand. Recent releases listed are v4.3.0 on 2026-08-11, v4.2.8 on 2026-08-03 and v4.2.3 on 2025-11-26, and the last push on main is dated 2026-09-13.
Where does GeekAI store generated files?
File management is delegated to 七牛云, 阿里云 OSS, 腾讯云 OSS or Minio. The README does not give bucket naming rules, key formats or signed URL expiry settings for any of them.
Official sources
Add this badge to your README
If you maintain this project, the badge below links readers to this analysis and shows its maintenance status from the daily GitHub snapshot. Paste the markdown into your README; add ?metric=license or ?metric=stars to the image URL for a different field.
[](https://hysenlabs.com/projects/yangjian102621-geekai)Star history chart for your README
[](https://hysenlabs.com/projects/yangjian102621-geekai)Was this useful?