How to Use Coze Variables and Long‑Term Memory in Low‑Code Agent Development (Part 6)

Coze Variables

Variables store user information and guide dialogue flow. Example: variable user_name records the name from the first interaction and can be reused to greet the user on subsequent visits. Variable user_visits increments on each visit to differentiate greetings for first‑time and returning users.

Setup Steps

Open Coze platform at https://www.coze.cn/home and create a new agent.

In the orchestration view, locate the “Variables” section and click the “+” button to add a variable.

Enter variable name, default value, and description, then save.

System preset variables such as sys_uuid, sys_longitude, sys_latitude are read‑only.

Create a variable visit_times to record the number of user visits.

Prompt Logic Example

If visit_times = 1, ask for the user’s name.

If visit_times > 1 and name exists, greet with “{{name}}, welcome back for the {{visit_times}}‑th time!”

If visit_times > 1 and name is missing, ask for the name again.

After each greeting, visit_times increments by 1.

# Role
You are a smart greeting assistant that greets based on visit count.

## Skills
### Skill 1: First‑visit greeting
1. When `visit_times` = 1, output: "你好，我是你的智能问候助手，请问怎么称呼你？"

### Skill 2: Non‑first‑visit greeting
1. When `visit_times` > 1 and `name` exists, output: "{{name}}，欢迎第{{visit_times}}次光临!很高兴再次见到你。"
2. When `visit_times` > 1 and `name` missing, output: "欢迎第{{visit_times}}次光临!我是你的智能问候助手，请问怎么称呼你？"

## Constraints
- After each greeting, `visit_times` increments by 1.
- Follow the given conditions and output format exactly.

Long‑Term Memory

Long‑term memory records every dialogue turn, extracts key information, and stores it as structured memory data. When a user returns, the agent can recall past interactions to generate coherent, personalized responses.

Enabling

For agents created before 2025‑10‑13, enable long‑term memory with a single click in the orchestration page. For newer agents, first create a memory store, then bind it in the agent configuration.

Retail Scenario

A user asks about a summer dress. The agent remembers the color preference, suggested size, and fabric‑care advice in the next session, replying:

“Welcome back! I remember last week we discussed the summer dress you liked—light‑blue French‑print. Based on your height and weight, I suggested size S, but you might prefer XS for a slimmer fit. Note the fabric should not be soaked long to avoid deformation…”

Combined Use

Variables act as a static user profile (e.g., name, age, gender, height, weight). Long‑term memory captures dynamic health data and conversation history. In a health‑management assistant, the agent extracts statements about appetite, sleep, and diet, stores them in memory, and later generates a personalized recommendation that references both variables and memory points.

{{user_name}}，根据您的年龄（{{user_age}}岁）和体重（{{user_weight}}kg），我建议从以下几个方面进行调整：

1. **改善睡眠质量**
   尽量在晚上11点前入睡，保证每晚睡眠不少于7小时。您之前提到过睡眠不足的问题，可以尝试睡前泡热水澡或进行呼吸放松练习。

2. **优化饮食结构**
   多吃富含维生素C和E的新鲜蔬果，减少高糖高脂摄入。关于之前讨论的减脂食谱，如果您在执行中遇到困难，我们可以进一步调整方案。

3. **加入适度运动**
   建议进行低强度有氧运动，如慢跑、游泳等，有助于促进新陈代谢、改善血液循环。

4. **针对性补充营养**
   可适当食用柑橘类水果，或考虑服用维生素C与E的补充剂，有助于改善皮肤状态。

如果您在执行过程中有任何不适或新的情况，欢迎随时告诉我，我将根据您的进展提供进一步建议。

Summary

Variables provide static user attributes; long‑term memory preserves evolving context across sessions. Together they enable personalized, multi‑turn interactions. Coze also supports integration with structured databases for richer data management.