​
About wiseflow

wiseflow (Chinese name: AI Chief Intelligence Officer) is an open-source application that leverages large language models to help users discover genuinely interesting information from vast amounts of data and diverse sources on a daily basis.

Unlike ‘deep search’ tools such as ChatGPT and Manus, wiseflow is designed for ‘wide search’, making it particularly suitable for scenarios that require broad information collection such as industry intelligence, customer insights, bidding information, competitor dynamics, public opinion monitoring, and knowledge intelligence. Compared to traditional RPA-based crawlers, the project also supports a plug-and-play experience that eliminates the need for manual XPath extraction, using large models to strictly analyze, filter, and summarize each piece of information according to user-defined focus points.

wiseflow has just released its brand-new version 4.0. In addition to existing sources like regular web pages, RSS, and search engines, it now supports platforms such as Weibo and Kuaishou. Future versions in the 4.x series will gradually add support for platforms including WeChat Official Accounts, Douyin, Xiaohongshu, Bilibili, and Zhihu. The 4.0 release also features a significant architectural overhaul, resulting in improved stability, speed, and memory efficiency.

​
How to Use

Since version 3.1, wiseflow has recommended using LLM services provided by SiliconFlow. Version 4.0 further simplifies the setup process—just three steps to get started!

Tip: Windows users should download the Git Bash tool in advance and run the following commands in Bash. Download Git Bash

​
Download the Project Source Code and Install uv and PocketBase

curl -LsSf https://astral.sh/uv/install.sh | sh
git clone https://github.com/TeamWiseFlow/wiseflow.git

The above steps will complete the installation of uv.

Next, go to the PocketBase documentation to download the PocketBase executable compatible with your operating system, and place it in the wiseflow/pb directory.

Note: If you’re using macOS, after completing this step, navigate to the wiseflow/pb folder and run the following command once:

xattr -d com.apple.quarantine pocketbase

​
Configure the .env File Based on env_sample

First, in the wiseflow directory (the project root), create a .env file based on the env_sample file and fill in the necessary configuration values.

Now it’s time for SiliconFlow to take the stage!

  • Visit the SiliconFlow official website and register an account (or log in if you already have one);
  • After registration, go to the API Key page and create a new API Key;
  • Click to copy the key, then paste it into the LLM_API_KEY field of the .env file you created earlier. Also set LLM_API_BASE to https://api.siliconflow.cn/v1.

A sample .env file might look like this:

LLM_API_KEY=your_api_key 
LLM_API_BASE=https://api.siliconflow.cn/v1 # LLM service endpoint
PRIMARY_MODEL=Qwen/Qwen3-14B # Recommended: Qwen3-14B or similar reasoning-capable models
VL_MODEL=Pro/Qwen/Qwen2.5-VL-7B-Instruct

For wide-range information gathering tasks like wiseflow’s, the Qwen3-14B model is already quite effective. The VL_MODEL is auxiliary—Qwen2.5-VL-7B-Instruct will also suffice. wiseflow is optimized for smaller models, but if your use case involves strict constraints and lots of proprietary terms, you may consider using a larger model like Qwen3-32B.

Refer to docs/manual/manual.md for more .env configuration options.

​
Launch!

Now just run the following commands in order:

cd wiseflow
uv venv # Only needed the first time
source .venv/bin/activate  # Linux/macOS# 
# Or on Windows:
# .venv\Scripts\activate
uv sync # Only needed the first time
python -m playwright install --with-deps chromium # Only needed the first time
chmod +x run.sh # Only needed the first time
./run.sh

The project will automatically launch a browser, guiding you through registration and into the user interface.