Building a Personal Blog Knowledge Base with Coze AI Bot Platform

Since the release of ChatGPT 3.5 in 2022, AI technology has rapidly expanded, prompting discussions about its impact on society, employment, privacy, and inequality. 2023 marked a pivotal year for large language models (LLMs) such as GPT‑4, Llama 2, and many Chinese equivalents, which have evolved from simple chat to multimodal capabilities like text‑to‑image and image‑to‑text.

Coze (扣子) is a next‑generation, one‑stop AI Bot development platform that allows users with or without programming experience to quickly build bots powered by LLMs. This article demonstrates how to leverage Coze to construct a personal blog knowledge base that can answer queries using the blog’s own content.

The functional architecture of Coze includes built‑in and custom plugins (e.g., search, data retrieval), multiple data sources (content upload, URL crawling, API JSON, custom datasets), persistence options (database and knowledge‑base storage), and a visual workflow engine that links these components.

In the workflow, the start node defines a query:string parameter. The search_keywords plugin node fetches blog data via an API and returns JSON. A code node then filters the results using the following script:

async function main({ params }: Args): Promise
{
  const regex = new RegExp(params.input, 'i');
  let ret = [];
  for (let item of params.search_result) {
    var data = {
      "content": "",
      "date": "",
      "permalink": "",
      "summary": "",
      "title": ""
    };
    if (regex.test(item.title)) {
      data.content = item.content;
      data.date = item.date;
      data.permalink = "https://fm126.top" + item.permalink;
      data.summary = item.summary;
      data.title = item.title;
      ret.push(data);
    }
  }
  return ret;
}

The end node returns the filtered array ( ret ) to the workflow output.

To populate the knowledge base, the blog’s pages are imported as “online data”. Coze automatically segments the uploaded content into discrete chunks, which are then indexed for vector‑based similarity search, enabling precise answer retrieval.

Creating the Bot involves defining a name, description, and a detailed prompt that sets the bot’s persona and response logic. Additional tool configurations include the Bing search plugin for fallback queries, and advanced settings such as opening messages, suggested questions, and voice tones.

After configuring the workflow, plugins, and knowledge base, the Bot can be previewed, debugged, and iteratively refined. Once performance meets expectations, the Bot is published to platforms like Feishu via Coze’s publishing guide.

Overall, Coze provides a convenient, low‑code solution for building a domain‑specific AI assistant, though careful prompt tuning and consistent testing are essential to achieve reliable results.