ChatWiki's deployment guide prints the default password, and its newest changelog is mostly about the hosted product
ChatWiki is an AI knowledge-base workflow agent platform for WeChat official accounts, offering a RAG-based AI customer-service chatbot, aiming to become a vertical-domain Coze or n8n.
At a glance
- What is it?
- zhimaAi/chatwiki is a workflow automation platform for the WeChat ecosystem, with RAG knowledge bases, human handoff and MCP integration, distributed as a Docker compose community edition. What the page gives you plainly is a default administrator password in the install instructions, a plugin system whose download URLs come from the server, and a release note that spends four of nine items on a cloud tier.
- Who is it for?
- Use ChatWiki if you are building a customer service or knowledge assistant inside the WeChat ecosystem and want the triggers, the fan tagging and the article drafting already wired, because those integrations are the reason the project exists. Three things to do before exposing anything.
- Can I use it commercially?
- Check first. The repository uses a licence we do not classify automatically, so read its LICENSE file before any commercial use.
- Is it still maintained?
- Yes. The repository last received commits 2 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
- zhimaAi/chatwiki2,114 ★ · +44
| Repository | 2026-09-10 | 2026-11-15 |
|---|---|---|
| zhimaAi/chatwiki | 2.1k | 2.1k |
44 Stars in the last 66 days, measured from the daily GitHub snapshot. Compare with other projects
Editorial analysis
The install guide publishes the default administrator password
The one-click deployment section is five commands, and the last two lines are the reason to read it carefully:
# Install Docker
sudo curl -sSL https://get.docker.com/ | CHANNEL=stable sh
# Clone the project
git clone https://github.com/zhimaAi/chatwiki.git
cd chatwiki/docker
# Start the service
docker compose up -d
# Start using it, access via IP:port (ensure the specified port ${CHAT_SERVICE_PORT}, default 18080, is open)
# Default username: admin
# Default password: chatwiki.com@123The password is in the documentation, which means it is the first thing anyone scanning a newly deployed instance will try. The port is the other half: the guide tells you to make sure the service port is open, and names 18080 as the default, so a first install on a cloud host is reachable at a predictable address with predictable credentials until somebody changes both.
The first line is also worth noting as a habit rather than a criticism of this project. It pipes a remote script into a shell with root privileges, which is how Docker's own convenience installer works, and it is the standard way to install Docker on a server. It is also the point at which the security of the whole stack rests on a text file served over HTTPS from a domain you did not choose.
The password string is a fair sample of the project's conventions: it embeds one of the project's own hostnames, so the default credential is itself a piece of branding.
Automatic replies to private messages on an unverified official account
The feature list leads with a claim it labels industry first, and it is the one that matters most for how the platform is used. ChatWiki can reply automatically to private messages for unverified official accounts, and it handles text, voice, images, mini-program cards and video messages rather than plain text alone.
An unverified official account is a WeChat account that has not completed the platform's verification process, and the platform's rules about what such an account may do are the reason this capability is worth a claim rather than a feature bullet. A tool that answers private messages on a user's behalf, with generated content, is an automated account, and that is exactly the behaviour platforms restrict. The project presents it as a capability and does not discuss the platform policy behind it, which means an adopter is the one who has to answer that question.
The trigger list is broader than messaging. Triggers include user private messages, comments, follows, unfollows and menu clicks, and the processing steps include replying to private messages, tagging fans, generating draft articles and publishing articles. Follow and unfollow as triggers is unusual, and combined with fan tagging it means the workflow can react to the shape of an audience rather than only to what someone said.
Plugin download URLs are fetched from the server, and a JavaScript interpreter is in the build
The newest changelog entry that concerns security is item five of the 2026-09-18 release, and it says that plugin download URLs are now retrieved from the server. It is filed as a system security optimisation, which tells you two things: the URLs were previously decided locally, and moving them server side was considered an improvement.
It is also the most consequential architectural fact on the page, because a plugin system that resolves its own download addresses is a system that fetches and runs code chosen by a remote host. The project does not say on this page what a plugin is written in, what executes it, or what verification happens after download.
The dependency list offers a clue without answering the question. Among the direct requirements is a JavaScript interpreter written in Go, which is the kind of dependency a plugin system pulls in when plugins are scripts rather than binaries. It sits next to an HTML template renderer, a Markdown parser, a PDF library and a set of article-scraping libraries, so the runtime already contains a document pipeline as well as a scripting engine.
The honest reading is that plugins are a real capability with a real supply chain, and the page documents one change to how that chain is addressed and nothing about what protects you after the download.
Authentication rests on a preview version, and the module list is very long
The module file names the module `chatwiki` and the language line is Go 1.25.0. The direct requirement list is where a careful reader should stop, because it is long and because two entries carry versions that ask for attention.
One is the JSON web token library, pinned to a version whose own version string calls it a preview. A preview dependency in an authentication path is worth knowing about, particularly in a product whose permission story otherwise looks considered, with a three-tier role system and IP whitelisting.
The other is the CloudWeGo model framework, pinned to a release whose version string marks it as a beta. For a project whose core function is orchestrating model calls, a beta framework dependency is the kind of thing that moves under you.
The rest of the list is a picture of what the product does. A WeChat API wrapper with its supporting power library, an authorization library paired with a database adapter, an object storage SDK and an Alibaba Cloud credentials library, a DingTalk client, a document intelligence API client, a Markdown parser, an HTML readability extractor with two query engines and an XML parser, a PDF toolkit, a Word document library, an image processing library, an SMTP client, a Redis client, a web framework, a configuration library, an internationalisation package, an environment variable loader and an SFTP client. That is a large surface for a self-hosted service, and the permission model has to hold across all of it.
Vue, Go, Python and a PHP directory in one repository
The declared primary language is Vue, and the tech stack section says the frontend is vue.js, the backend is golang plus python, and the database is PostgreSQL 16 with pgvector and zhparser. That last entry is a detail rather than a detail: a Chinese word segmenter alongside a vector extension means the retrieval path is a hybrid one, with full text search and vectors over Chinese text rather than English stemming.
The tree then adds a third and fourth language to the picture. Alongside the Go module, the front-end directory, the command and internal Go packages, the configuration and Docker directories, there is a `php/` directory, an `html-template/` directory, a `clawbot/` directory, a `volumes/` directory checked into the repository, and a Windows batch file named `zig-linux-cc.cmd`, which is a Zig based cross-compilation helper for Linux binaries.
So the repository is a polyglot deployment rather than a single application, and a reader trying to answer a simple question, which service do I patch, has four plausible answers. The committed volumes directory is the one to check before running anything, since data directories inside a repository are either fixtures or leftovers, and neither is a good thing to discover after a first deploy.
Nine release items, four of them about the hosted tier
The v2.9.4 release and the last push share a timestamp to the second, so the tag corresponds to that changelog block. It has nine items and they are worth reading as a description of where the product's attention is.
Four items are explicitly about the cloud version. Intent-based human handoff gained custom recognition criteria, credit usage statistics were added, a bug where the trigger count in handoff keyword statistics ignored the selected time range was fixed, and the VIP system gained a flagship edition. A fifth item caps the number of workflow calls in the free cloud version.
The remaining four are closer to the code. Plugin download URLs moved server side, the timeout in the workflow HTTP request node was adjusted, a bug storing the wrong model names in custom model statistics was fixed, and the knowledge base retrieval test page now remembers what you typed. The earlier block, dated 2026-09-04, adds a recycle bin to the Q and A knowledge base, raises the sensitive word limit to 20,000, updates an external service page built on Alibaba Cloud, and fixes a case where context was not passed in some situations.
So the headline release of the open source repository is substantially a hosted product changelog, with a pricing tier and a usage cap in it. Anyone self-hosting the community edition is reading about features they will not find in the code they just cloned.
Three hostnames, and the documentation lives on a third-party wiki
The project's own naming is inconsistent in a way that matters for anyone about to type a URL. The homepage recorded for the repository is one domain with a dot-ai ending. The banner at the top of the readme points at the dot-com version. The free trial is given as the dot-com version again. And the default installation password contains the dot-com hostname as a string.
Alongside that, the documentation is split across three places. The readme carries the feature list and the deployment commands, a separate file in the repository root holds the changelog, a Chinese version of the readme sits beside the English one, and the actual guides are hosted on a third-party wiki with a dozen entries: installation through an assistant, one-click deployment, mirror and offline installation, deployment without Docker, deployment on two different Chinese control panels, how to configure model providers, local model deployment, push notification configuration, and how to obtain model API keys.
The help documentation links carry a query parameter with a base64-encoded URL appended to them, which is a tracking convention rather than a technical one, but it does mean the links you click are not the links in the source.
Contact is by email at an address on a different domain from both the product and the documentation, plus a WeChat group. That spread of destinations is common in Chinese infrastructure products and it is also the reason a self-hosted evaluation takes longer here than the deployment commands suggest.
Editorial conclusion
Use ChatWiki if you are building a customer service or knowledge assistant inside the WeChat ecosystem and want the triggers, the fan tagging and the article drafting already wired, because those integrations are the reason the project exists. Three things to do before exposing anything. Change the default administrator credentials, which the deployment guide prints in plain text, and do not leave the service port reachable from the internet before you do. Decide what the plugin system is allowed to fetch, since plugin download URLs are now retrieved from the server and a JavaScript interpreter sits in the dependency list, so the plugin surface is code arriving from a remote host. And decide which edition you are running, because most of the newest changelog items are marked as cloud version features, including a paid flagship tier and a cap on free workflow calls, which means the open source community edition and the hosted product are not the same product. Support is documented on a third-party wiki rather than in the repository.
Frequently asked questions
What is the zhimaAi ChatWiki platform?
A workflow automation platform focused on the WeChat ecosystem, with a document knowledge base, a Q and A knowledge base, retrieval augmented generation, human handoff and MCP integration. It supports more than 20 models including DeepSeek R1, doubao pro, qwen max, OpenAI and Claude, and its own description positions it as a vertical-domain alternative to Coze and n8n.
How do I deploy the ChatWiki Community Edition?
With Docker: install Docker, run `git clone https://github.com/zhimaAi/chatwiki.git`, change into the `chatwiki/docker` directory and run `docker compose up -d`. You then reach it over IP and the port given by `CHAT_SERVICE_PORT`, which defaults to 18080 and has to be open for the service to be reachable.
What are the default ChatWiki login credentials?
The deployment instructions print a default username of admin and a default password of chatwiki.com@123. They also tell you to make sure the service port is open, so a fresh installation is reachable on a predictable address until both the credentials and the network exposure are changed.
What can a ChatWiki workflow do inside WeChat?
Triggers include user private messages, comments, follows, unfollows and menu clicks, and the processing steps include replying to private messages, tagging fans, generating draft articles and publishing articles. It also claims automatic replies to private messages on unverified official accounts, handling text, voice, images, mini-program cards and video messages.
What knowledge bases does ChatWiki support?
Two. A document knowledge base with URL reading, batch import, API integration, several segmentation strategies, knowledge graphs and hybrid vector search, and a Q and A knowledge base that extracts question and answer pairs from uploaded documents, clusters unknown questions automatically and summarises frequent questions from real conversations.
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/zhimaai-chatwiki)Star history chart for your README
[](https://hysenlabs.com/projects/zhimaai-chatwiki)Was this useful?