# PandaWiki: a Docker-only AI knowledge base from Chaitin > PandaWiki pairs a rich-text wiki with AI writing, Q&A and search, and installs through a single shell script on a Docker host. The catch is that the whole AI layer is inert until you connect a model provider. **chaitin/PandaWiki** — PandaWiki 是一款 AI 大模型驱动的开源知识库搭建系统，帮助你快速构建智能化的 产品文档、技术文档、FAQ、博客系统，借助大模型的力量为你提供 AI 创作、AI 问答、AI 搜索等能力。. - Repository: https://github.com/chaitin/PandaWiki - Website: https://pandawiki.docs.baizhi.cloud/ - Stars: 10,292 · Forks: 1,024 - Language: TypeScript - License: AGPL-3.0 - Published: 2026-08-04 · Updated: 2026-08-18 · Language: en - Canonical page: https://hysenlabs.com/projects/chaitin-pandawiki ## What PandaWiki is for, and who it is aimed at PandaWiki is an open source knowledge base system driven by large language models, written mainly in TypeScript and published by Chaitin under AGPL-3.0. The README frames it as a way to build product documentation, technical documentation, FAQ pages and blog systems, with AI assistance layered on top for writing, question answering and search. The unit of organisation is the knowledge base: a collection of documents from which PandaWiki generates a separate Wiki site. So a single installation can host several distinct public-facing sites, each backed by its own document set. The intended user is a team that already has documentation to publish and wants a chat-style interface over it without assembling a retrieval stack by hand. The README lists integrations that point the same way: a web widget you can embed in another site, and chat bots for DingTalk, Feishu and WeCom. Content can also arrive from outside, by web page URL, by site sitemap, by RSS feed, or by offline file import. That import surface matters more than the editor for anyone migrating an existing docs folder. It is not a general-purpose CMS. There is no mention of themes, plugins, or a template marketplace in the README, and the feature list is short and specific. If you need a marketing site with a blog and a landing page builder, this is the wrong shape of tool. ## How the pieces fit: console, Wiki sites and a model you supply The architecture visible from the repository is split into a backend, a web frontend, an sdk directory and a set of images. The runtime picture from the README is simpler: you get a console for managing knowledge bases and uploading documents, and separately a Wiki site that readers use. The console is where documents are uploaded and where you wait for them to be learned. The Wiki site is where AI Q&A is exercised against them. The dependency that shapes everything else is the model. The README states plainly that PandaWiki is a model-driven wiki system and that without a configured model, AI creation, AI Q&A and AI search will not work. On first login the console prompts for model configuration, offering either a one-click automatic setup or manual configuration. The README recommends Baizhi Cloud's model marketplace for a quick connection and notes a five yuan credit on registration, which is a vendor recommendation rather than a technical requirement. That means the retrieval behaviour is not something you can inspect from the repository alone. The README does not describe the embedding model, the chunking strategy, the vector store, or how documents are indexed after upload. You know documents are learned and that answers come back, and nothing about the pipeline between those two points. If retrieval quality is the deciding factor for you, the documentation is silent and you would need to read the backend source. ## Installing PandaWiki on a Docker host The README gives one supported path: a Linux system with Docker 20.x or newer, logged in as root. The installer is a shell script fetched and executed in one line. It is interactive, and the README warns the process takes several minutes. ```bash bash -c "$(curl -fsSLk https://release.baizhi.cloud/panda-wiki/manager.sh)" ``` When it finishes, the terminal prints a success block containing internal and external access addresses on port 2443, a username, and a generated password. Open one of those addresses in a browser to reach the console login. The README's example output shows the same port for both the internal and external address. ``` SUCCESS 控制台信息: SUCCESS 访问地址(内网): http://*.*.*.*:2443 SUCCESS 访问地址(外网): http://*.*.*.*:2443 SUCCESS 用户名: admin SUCCESS 密码: ********************** ``` After logging in, the console asks you to configure an AI model before anything else, with a one-click option or manual entry. With a model in place, create a knowledge base, upload documents, wait for them to be learned, then open the Wiki site to test Q&A. The README does not document how to change the console port, how to run the installer non-interactively, or how to roll back an installation. ## Two constraints the README does not soften The first is the host requirement. Installation is documented only for Linux with Docker 20.x or newer, run as root. There is no Windows or macOS install path in the README, and no mention of a hosted option. If your team standardises on Windows workstations, you are running a Linux VM or a remote server, not installing locally. The related searches for Windows and Mac versions have no corresponding instructions in the repository. The second is the model dependency, and it is a real failure mode rather than a setup detail. A fresh installation is a wiki with no AI. Upload documents, open the Wiki site, and the Q&A and search features are unavailable until a model is configured. That also means every deployment carries an external dependency: an API key, a provider account, and whatever cost and rate limits come with it. The README's recommended route is a paid marketplace with a small trial credit, so a self-hosted PandaWiki is not a fully offline system out of the box. A third, quieter limitation is the licence. AGPL-3.0 requires that if you run modified PandaWiki as a network service, you release your modifications under the same licence. The README states this directly, including that commercial use carries the same open source obligation. For an internal docs portal that is usually fine. For a product you intend to sell as a hosted service on top of modified code, it is a design constraint you should settle before writing code, and it is not a question this article can answer for your situation. ## PandaWiki compared with MaxKB, RAGFlow and QAnything The alternatives people search for alongside PandaWiki are MaxKB, RAGFlow and QAnything, and the difference is one of starting point rather than features. Those three are retrieval and question-answering engines: you bring documents, they build a pipeline, and the interface is a chat or an API. PandaWiki starts from the publishing side. Its primary artifact is a Wiki site that readers browse, with AI Q&A attached to it, plus a rich-text editor that handles Markdown and HTML and exports to Word, PDF and Markdown. That ordering changes what you spend time on. With a RAG engine, the work is tuning retrieval, chunk sizes, and evaluation. With PandaWiki, the work is writing and organising documents, and the AI behaviour is largely delegated to the model you configure, since the README documents no retrieval tuning knobs at all. If your problem is answer quality over a messy corpus, a dedicated engine gives you more surface to adjust. If your problem is that you have documentation to publish and want a chat box on it, PandaWiki covers both halves in one install. The export and integration features are the practical tiebreaker. Word and PDF export, an embeddable widget, and DingTalk, Feishu and WeCom bots are documented here; the same list is not what those retrieval engines lead with. ## Maintenance, releases and upgrade cost The repository is not archived, and the last push was on 2026-08-21, which is under a month before this article. Release tags in the provided history are v3.87.0 on 2026-08-21, v3.86.5 on 2026-08-13 and v3.86.4 on 2026-07-23. The cadence in that window is frequent, and the version numbers move in small increments, which suggests a steady stream of patch releases rather than rare large ones. What the README does not give you is an upgrade procedure. It covers installation and first login, and points to external documentation for deployment details, but it does not describe how to move an existing installation from v3.86.5 to v3.87.0, whether data survives an upgrade, or how to back up a knowledge base before trying. That is the main operational unknown. A project releasing this often is one you will want to update, and the repository as presented does not tell you how to do it safely. On licence cost: AGPL-3.0 imposes no fee, but it does impose source disclosure for modified network services, as the README notes. The other recurring cost is the model provider. Nothing in the README suggests PandaWiki ships its own model, so budget for API usage separately from hosting. ## Conclusion Adopt PandaWiki if you want a self-hosted wiki whose Q&A and search ride on your own model credentials, and you are comfortable running Docker on Linux and meeting AGPL-3.0's network-copyleft terms. Skip it if you need a Windows or macOS host, or if you want a knowledge base that works without wiring up an external model. Before committing, verify three things: that your host runs Docker 20.x or newer, that the console port 2443 is reachable, and that you have an API key for whichever model provider you intend to use, because AI creation, Q&A and search stay unavailable until a model is configured. ## FAQ ### What are the system requirements for installing PandaWiki? The README specifies a Linux system with Docker 20.x or newer, and the install command is run with root privileges. No Windows or macOS installation path is documented. ### Does PandaWiki work without configuring an AI model? No. The README states that AI creation, AI Q&A and AI search cannot be used normally until a model is configured, and the console prompts for model setup on first login. ### Which port does the PandaWiki console listen on after installation? The installer's success output shows the console access address on port 2443 for both the internal and external address. The README does not document changing that port. ### What licence does PandaWiki use and what does it require? PandaWiki is licensed under AGPL-3.0. The README states that modifications must be open sourced under the same licence, that providing the software as a network service also requires releasing your code, and that commercial use follows the same open source requirements. ## Sources - [Official documentation](https://pandawiki.docs.baizhi.cloud/) - [Official README](https://github.com/chaitin/PandaWiki#readme) - [Project repository](https://github.com/chaitin/PandaWiki) - [Release notes](https://github.com/chaitin/PandaWiki/releases) --- Hysen Labs editorial analysis, written from the project's own repository and release notes. Cite the canonical page: https://hysenlabs.com/projects/chaitin-pandawiki