简体中文 | English

only ONE executable file, you can use it directly, including intent detection, AI management, a visual process editor and a response system.
ZERO installation, no need to install any middleware such as Redis, Elastic Search, etc.

• 🛒 Light Only ONE executable file, it can run smoothly on laptops without GPUs (data files will be created at runtime automatically).
• 🐱‍🏍 AI powered Integrated Huggingface local models (Llama, Phi-3, Gemma, Multilingual E5, MiniLM L6v2, NomicEmbedTextV1_5, etc.), and any OpenAI-compatible endpoint — OpenAI, DeepSeek, Zhipu GLM, Qwen, Moonshot Kimi, SiliconFlow, Groq, OpenRouter, Ollama, vLLM, LM Studio, or your own gateway. Just fill in the URL and API key. This can be used for Chat, Text generation and Intent detection.
• 🚀 Fast Built on Rust and Vue3.
• 😀 Simple Use the mouse to drag and drop with our intuitive node-based editor.
• 🔐 Safe 100% open source, all runtime data is saved locally (Using OpenAI API may expose some data).

• 🐋 Docker We provided an image on Docker Hub at dialogflowai/dialogflow
• 💻 Binary releases, please check here

By default application will listen to 127.0.0.1:12715, you can use -ip and -port specify new value, e.g.: dialogflow -ip 0.0.0.0 -port 8888

POST /flow/answer returns the answer in one document unless the request body asks for "stream": true. Then the response is application/x-ndjson: one JSON frame per line, each carrying a piece of an answer ({"contentSeq": 0, "content": "..."}), and the last line is always a terminal frame (contentSeq: null) whose content is the whole {status, data, err} response, so the final nextAction and collectData arrive with it.

curl -N -H 'Content-Type: application/json' \
  -d '{"robotId":"...","mainFlowId":"...","userInput":"hi","stream":true}' \
  http://127.0.0.1:12715/flow/answer

A client that does not send stream gets exactly the JSON it got before 1.23. See doc/streaming.md for the full protocol, the Java/JavaScript client contracts and the known limitations.

https://dialogflowai.github.io/

Node	Name
	Dialog Node
	Large language model chat node
	Knowledge base answer node
	Conditions node
	Goto node
	Collect node
	External HTTP node
	Send email node
	The end node

Using the different nodes above, to arrange and combine, you can get a conversational bot that can handle problems in different scenarios.

1. docker pull dialogflowai/demo
2. docker run -dp 127.0.0.1:12715:12715 --name dialogflowdemo dialogflowai/demo
3. Open your browser and visit: http://127.0.0.1:12715/

1. From Github release page, depending on the operating system, download the application.
2. Run it directly, or use the -ip and -port parameters to perform the listening IP address and port, e.g.: dialogflow -ip 0.0.0.0 -port 8888.
3. Open your browser and visit http://localhost:12715 (by default) or http://new IP:new port to see the application in action
4. Add a main flow and click its name into it
5. Create dialog flow by dragging and drop nodes onto canvas
6. Test it