ChatWiki: WeChat-Native RAG and Workflow Platform for AI Customer Service Agents

What ChatWiki Is Built For

WeChat official accounts are the primary customer communication channel for a large share of Chinese businesses. They allow businesses to send messages, manage followers, respond to comments, and automate interactions through the official account platform API. Building an AI-powered customer service layer on top of a WeChat official account typically requires custom development against that API.

ChatWiki describes its positioning as a workflow automation platform focused on the WeChat ecosystem, dedicated to making every official account a super AI agent. It handles the WeChat API integration, provides a visual workflow builder for defining how the agent responds to different triggers, and layers RAG knowledge bases on top so the agent can answer product questions from stored documents.

The README identifies the project as targeting the same market as Coze and n8n but specialized for WeChat and vertical-domain AI customer service. The product is aimed at businesses and developers who want to automate WeChat official account interactions without writing and maintaining custom API code.

WeChat Ecosystem Integration in Detail

The README describes several WeChat-specific capabilities that are not standard in general RAG platforms. Automatic reply to private messages works for unverified official accounts, covering text, voice, images, mini-program cards, and video messages. This is described as an industry first in the README.

The WeChat workflow triggers include: user private messages, user comments, follows, unfollows, menu clicks, and more. Supported processing steps include replying to private messages, tagging fans, generating draft articles, and publishing articles. This means the agent can not only answer customer questions but also perform administrative actions on the official account itself.

Knowledge base synchronization allows scraping articles and materials from WeChat official accounts directly into the knowledge base with one click, turning existing published content into retrieval sources without manual export and import.

The platform also supports publishing agents to WeChat service accounts, WeChat Work (enterprise accounts), and other WeChat-adjacent channels including website embedding and a standalone WebApp.

Deploying ChatWiki with Docker

ChatWiki Community Edition deploys via Docker Compose:

sudo curl -sSL https://get.docker.com/ | CHANNEL=stable sh
git clone https://github.com/zhimaAi/chatwiki.git
cd chatwiki/docker
docker compose up -d

The default port is 18080 (configurable via the CHAT_SERVICE_PORT environment variable). The default credentials are username admin and password chatwiki.com@123.

The tech stack is Go for the backend, Python for ML components, Vue.js for the frontend, PostgreSQL 16 with pgvector for vector storage, and zhparser for Chinese text segmentation. The combination of pgvector and zhparser is specific to Chinese-language RAG pipelines; zhparser enables accurate tokenization of Chinese text for retrieval.

Alternative deployment guides in the README cover the Baota Linux Panel, 1Panel, offline Docker installation, deployment without Docker, and local model deployment. The help documentation is hosted on Yuque and covers model provider configuration, API key acquisition, and domain setup for WeChat push notifications.

Knowledge Base Types and Workflow Orchestration

ChatWiki supports three knowledge base types. Document knowledge bases support URL reading, batch document import, API integration, and multiple segmentation strategies: AI-based segmentation, QA-based segmentation, and parent-child segmentation. The retrieval layer uses hybrid vector search combined with knowledge graph exploration, and the platform includes a visual knowledge graph explorer. Parent-child segmentation maintains a hierarchy between document chunks, so retrieval can return the parent context when a child chunk matches, improving coherence for long-form documents.

QA knowledge bases extract question-answer pairs from uploaded documents automatically, support clustering of unknown questions across live conversations, and summarize common FAQs from human agent conversation histories. This type is particularly useful for FAQ automation where the source material is structured as question-answer pairs. The clustering of unknown questions is a practical feature for discovering gaps in coverage: questions the bot cannot answer accumulate in a review queue, and operators can use them to identify which topics need additional knowledge base entries.

The workflow orchestration system supports conversational workflows and plugin workflows. Nodes include standard workflow logic, bidirectional MCP (Model Context Protocol) integration, agent mode, and user interaction nodes. MCP integration allows connecting to external MCP services or publishing workflows as MCP services. The complete OpenAPI interface allows integration with external business systems so workflows can be triggered from or write results to existing CRM or ticketing systems.

The model support list covers over 20 providers including DeepSeek R1, doubao pro, qwen max, OpenAI, and Claude. The go.mod shows dependencies on the CloudWeGo eino framework for LLM orchestration, which is a Go-native LLM application framework developed at ByteDance.

Limitations of the WeChat-Centric Design

ChatWiki's deepest features are WeChat-specific. The comment reply automation, fan tagging, article publishing, and official account sync are only useful to teams operating WeChat official accounts. For teams building AI customer service for Slack, Telegram, email, or web chat, these features are irrelevant and the platform does not document integrations with those channels as primary use cases.

The platform requires PostgreSQL 16 with pgvector and zhparser. These are non-standard PostgreSQL extensions. zhparser is specifically for Chinese text segmentation; its inclusion means the platform is tuned for Chinese-language content. English or multilingual RAG pipelines may need different tokenization configuration that the README does not address in detail. The hybrid vector search is optimized for Chinese query patterns.

The licence type is listed as NOASSERTION in GitHub metadata, meaning it could not be automatically classified. The LICENSE file in the repository must be read before any production or commercial deployment. The project's WeChat and official account integration relies on API access granted by Tencent, which has its own terms of service for official account platform usage that apply independently of the ChatWiki code licence.

Human handoff is supported but requires configuration in the agent settings. The README notes that issues the bot cannot resolve escalate to human agents with multi-agent collaborative assignment, but the full configuration for that workflow is in the external Yuque documentation rather than the README. The cloud version and the self-hosted version differ in some features; the changelog uses [STD] to mark cloud-only items.

How ChatWiki Compares to Dify

Dify is a well-known open-source LLM application development platform. It provides a visual workflow builder, RAG document knowledge bases, agent mode, API access, and deployment options including cloud and self-hosted. Dify is channel-agnostic; it exposes an API that developers integrate with whatever chat interface or platform they are using.

ChatWiki focuses specifically on the WeChat ecosystem. Its WeChat trigger system, fan management actions, official account article sync, and direct publishing integrations are not available in Dify. For teams that primarily serve users through WeChat official accounts, ChatWiki's native integration removes significant custom development work.

For teams outside the WeChat ecosystem, Dify's broader channel support and larger community are more practical. ChatWiki's advantage disappears entirely when the target channel is not WeChat. For teams that already use Dify and want to extend into WeChat, the two platforms are not directly interchangeable; the WeChat trigger model in ChatWiki is not a plugin that can be added to a Dify installation.

Tech Stack, Changelog, and Maintenance

The tech stack is Vue.js for the frontend, Go and Python for the backend, and PostgreSQL 16 with pgvector plus zhparser for the database. The go.mod specifies Go 1.25.0 and includes notable dependencies: the CloudWeGo eino LLM orchestration framework, Casbin for access control, PowerWeChat for the official WeChat API client, and dop251/goja for JavaScript evaluation.

Version v2.9.4 was released on 2026-09-18. The changelog entry for that release includes human handoff improvements, credit usage statistics, VIP tier additions, security optimizations for plugin downloads, and workflow HTTP node timeout adjustments. The release cadence shown in UpdateLog.md is approximately two to three releases per month.

The last push to the main branch was on 2026-09-18. Community support is available via WeChat group and email at [email protected].

Editorial conclusion

ChatWiki is a strong fit for teams running WeChat official accounts who want to build AI-powered customer service on their existing audience without redirecting users to a separate channel. The Docker deployment is straightforward, the knowledge base tooling covers document, QA, and knowledge graph retrieval, and the workflow engine handles complex WeChat trigger scenarios. Teams outside the WeChat ecosystem will find the WeChat-specific triggers and sync features irrelevant and should consider Dify or a more general-purpose RAG platform instead. The licence type is not a standard SPDX identifier; read the LICENSE file before any commercial deployment. The last push was on 2026-09-18 with release v2.9.4, and the active release cadence is approximately two to three releases per month.

Frequently asked questions

Official sources

1. Issues
2. Project website
3. README
4. Releases
5. zhimaAi/chatwiki on GitHub