Skip to content

Commit e64bb7b

Browse files
author
harvey_xiang
committed
fix: confilct
2 parents e7ff1c2 + 8737088 commit e64bb7b

4 files changed

Lines changed: 295 additions & 0 deletions

File tree

‎content/cn/changelog.yml‎

Lines changed: 19 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -1,4 +1,23 @@
11
versions:
2+
- name: v2.0.4
3+
date: 2026-01-30
4+
changedInfo:
5+
New Features:
6+
- type: 核心能力
7+
changedInfo:
8+
- 新增 Skill Memory。
9+
- 新增对记忆版本功能做支持的相关模块。
10+
- 新增 chat/completions 接口，兼容 OAI 协议。
11+
Improvements:
12+
- type: 检索优化
13+
changedInfo:
14+
- 减少了检索阶段事实记忆与偏好记忆的重复。
15+
Bug Fixes:
16+
- type: 问题修复
17+
changedInfo:
18+
- 解决 Playground 新用户 bug。
19+
- 修复云服务添加记忆时偶发的卡顿问题。
20+
- 解决图数据库效率问题。
221
- name: v2.0.3
322
date: 2026-01-22
423
changedInfo:

‎content/en/changelog.yml‎

Lines changed: 19 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -1,4 +1,23 @@
11
versions:
2+
- name: v2.0.4
3+
date: 2026-01-30
4+
changedInfo:
5+
New Features:
6+
- type: Core Capabilities
7+
changedInfo:
8+
- Added Skill Memory.
9+
- Added modules supporting memory versioning.
10+
- Added chat/completions interface, compatible with OAI protocol.
11+
Improvements:
12+
- type: Retrieval Optimization
13+
changedInfo:
14+
- Reduced duplication between factual and preference memories during retrieval.
15+
Bug Fixes:
16+
- type: Bug Fixes
17+
changedInfo:
18+
- Fixed bug for new users in Playground.
19+
- Fixed occasional latency when adding memories in cloud service.
20+
- Resolved graph database efficiency issues.
221
- name: v2.0.3
322
date: 2026-01-22
423
changedInfo:

‎content/en/memos_cloud/features/advanced/skill.md‎

Lines changed: 256 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -0,0 +1,256 @@
1+
---
2+
title: Skills
3+
desc: Generate reusable Skill files for Agents by adding user conversation messages.
4+
---
5+
6+
## 1. What are MemOS Skills?
7+
8+
**Skills** are **modular capability packages** that an Agent can dynamically invoke while executing tasks. They are automatically dispatched and injected by the Agent based on the conversation context, requiring no manual intervention from the user. These skills are typically built by developers collaborating with LLMs, based on open-source projects or original concepts, and are continuously optimized through actual use.
9+
10+
MemOS advocates that "Memory is Asset." We believe that the resolution paths and user preferences precipitated in real conversations are essentially the most valuable materials for skills. Based on this philosophy, MemOS **now supports automatically extracting skills from user memories**—solidifying fragmented interaction histories into reusable, personalized professional capabilities.
11+
12+
::note
13+
**How are MemOS Skills different from existing memories?**
14+
15+
* **Static Facts → Dynamic Execution**
16+
17+
Memories are usually static and factual, such as: "I live in Shanghai" or "I like concise replies." This information provides the necessary context for Agent reasoning.
18+
19+
Skills are executable behavioral capabilities built upon memory. They encapsulate a clear set of task-processing logic, such as "How to plan a complete travel itinerary," guiding the Agent's decisions and actions.
20+
21+
* **Fragmented → Structured**
22+
23+
Memories are often fragmented, with each entry describing a single fact or preference.
24+
25+
Skills are highly structured, integrating multiple related memories into a complete task solution that can be reused across different tasks.
26+
::
27+
28+
## 2. How It Works
29+
30+
![image.png](https://cdn.memtensor.com.cn/img/1769759436251_3tx57c_compressed.png)
31+
32+
The diagram above illustrates the complete interaction flow between the end-user, your AI Agent, and MemOS:
33+
34+
1. Call the `add/message` interface to pass the user's conversation messages into MemOS.
35+
36+
2. Upon receiving the request, MemOS processes it in sequence to generate a Skill file:
37+
38+
a. **Intelligent Slicing**: Identifies task boundaries in historical conversations and slices them into task text blocks.
39+
40+
b. **Clustering & Extraction**: Clusters similar task blocks and combines them with the user's historical memory to extract structured skill text.
41+
42+
c. **Skill Transformation**: Converts the skill into an executable and recognizable Skill file.
43+
44+
3. Call the `search/memory` interface to retrieve memories. MemOS will return a unified result including context-related user facts, preferences, tool memories, and matching Skill files.
45+
46+
4. Download the Skill file and pass both the memories and the Skill file to your self-deployed LLM, enabling the effective utilization of long-term experience and automatically generated skills.
47+
48+
## 3. Usage Example
49+
50+
The following demonstrates an example of MemOS generating a "Travel Planning" skill based on historical conversations.
51+
52+
### 1. **Add Messages**
53+
54+
Add a conversation between a "High-Energy J-type person" and a "Travel Planning Assistant." The user expresses several requirements for the trip:
55+
56+
* Dislikes backtracking; "Special Forces" style (intensive) travel.
57+
58+
* Prefers cultural attractions.
59+
60+
* Needs to confirm weather and temperature in advance.
61+
62+
```python
63+
import os
64+
import requests
65+
import json
66+
67+
# Replace with your API Key
68+
os.environ["MEMOS_API_KEY"] = "YOUR_API_KEY"
69+
os.environ["MEMOS_BASE_URL"] = "https://memos.memtensor.cn/api/openmem/v1"
70+
71+
data = {
72+
"user_id": "memos_user_123",
73+
"conversation_id": "0127",
74+
"messages": [
75+
{"role": "user", "content": "I'm going to Chengdu next week. Help me plan a 5-day trip. I like 'Special Forces' style travel without backtracking. Please mark delicious local food along the way."},
76+
{"role": "assistant", "content": "...omitted..."},
77+
{"role": "user", "content": "I prefer visiting cultural sites; I'm not very interested in shopping malls."},
78+
{"role": "assistant", "content": "...omitted..."},
79+
{"role": "user", "content": "When planning, please check the weather and temperature in advance so I can pack my luggage."},
80+
{"role": "assistant", "content": "...omitted..."}
81+
]
82+
}
83+
headers = {
84+
"Content-Type": "application/json",
85+
"Authorization": f"Token {os.environ['MEMOS_API_KEY']}"
86+
}
87+
url = f"{os.environ['MEMOS_BASE_URL']}/add/message"
88+
89+
res = requests.post(url=url, headers=headers, data=json.dumps(data))
90+
91+
print(f"result: {res.json()}")
92+
93+
```
94+
95+
### 2. **Retrieve Memory**
96+
97+
Suppose the user makes another travel planning request. Pass the user's query and enable skill recall:
98+
99+
```python
100+
import os
101+
import requests
102+
import json
103+
104+
# Replace with your API Key
105+
os.environ["MEMOS_API_KEY"] = "YOUR_API_KEY"
106+
os.environ["MEMOS_BASE_URL"] = "https://memos.memtensor.cn/api/openmem/v1"
107+
data = {
108+
"query": "I plan to go to Yunnan during the Qingming Festival. Help me plan a 7-day itinerary.",
109+
"user_id": "memos_user_123",
110+
"conversation_id": "0301",
111+
"include_skill": True # Enable skill
112+
}
113+
headers = {
114+
"Content-Type": "application/json",
115+
"Authorization": f"Token {os.environ['MEMOS_API_KEY']}"
116+
}
117+
url = f"{os.environ['MEMOS_BASE_URL']}/search/memory"
118+
119+
res = requests.post(url=url, headers=headers, data=json.dumps(data))
120+
121+
print(f"result: {res.json()}")
122+
123+
```
124+
125+
### 3. **Result Display**
126+
127+
In the retrieval results below, the skill includes:
128+
129+
* Planning "Special Forces" itineraries.
130+
* Recommending cultural attractions.
131+
* Focusing on weather/temperature and recommending suitable clothing.
132+
133+
```markdown
134+
# Below is the Skill.md generated by MemOS
135+
136+
---
137+
name: Travel Itinerary Planning
138+
description: Design multi-day itineraries for travelers, including attraction arrangements, transportation, and weather adaptation suggestions.
139+
---
140+
141+
## Procedure
142+
1. Determine the traveler's interests and preferences 2. Gather information on attractions and activities at the destination 3. Plan the daily itinerary, ensuring high efficiency and no backtracking 4. Add local food recommendations to enrich the experience 5. Provide transportation and accommodation suggestions, balancing convenience and comfort 6. Check weather forecasts, adjust the itinerary, and prepare luggage
143+
144+
## Experience
145+
1. Efficient route design reduces commute time
146+
2. Prioritize attractions; avoid overly commercialized places
147+
3. Food recommendations increase the richness of the experience
148+
4. Weather adaptation ensures comfortable travel
149+
150+
## User Preferences
151+
- Itinerary arrangements without backtracking
152+
- Priority for cultural attractions, focusing on history and cultural experiences
153+
- Adjust itinerary and prepare luggage based on weather
154+
155+
## Examples
156+
157+
### Example 1
158+
159+
# Travel Itinerary Planning Example
160+
## Day 1
161+
- **Itinerary**: Panda Base → Eastern Suburb Memory → Jianshe Road Food Street
162+
- **Weather Adaptation**: Cloudy and windy, suitable for visiting cultural districts
163+
- **Food Recommendation**: Juntun Guokui (Pot-baked pancake), Soy milk
164+
- **Transportation**: Subway + Walking
165+
166+
## Day 2
167+
- **Itinerary**: People's Park → Chengdu Museum → Wenshu Monastery
168+
- **Weather Adaptation**: Light rain, focus on indoor museums
169+
- **Food Recommendation**: Zhong Dumplings, Chen Mapo Tofu
170+
- **Transportation**: Subway + Ride-hailing
171+
172+
...
173+
174+
## Additional Information
175+
176+
### Luggage Preparation Guide
177+
Based on destination weather characteristics, use the "onion layering" method for easy adjustment.
178+
179+
### Cultural Attraction Reservation Guide
180+
Provide reservation channels, ticket prices, and opening hours.
181+
182+
```
183+
184+
::note
185+
**Two Ways to Use Skills**
186+
187+
* If the Model/Agent you call has the capability to use Skill files, you can directly download the file from the `skill_url` link.
188+
* If the Model/Agent you call does not have the capability to use Skill files, you can directly convert `skill_value` into a string and add it to the prompt.
189+
::
190+
191+
### 4. **Build Your Personalized Skills**
192+
193+
Based on different users' conversation messages, MemOS can create skills exclusive to individuals. For instance, we constructed another conversation between a "Low-Energy P-type person" and a "Travel Planning Assistant." When they requested:
194+
195+
* Night owl, can't get up early.
196+
* Doesn't want to go to far-off places that require rushing.
197+
* Interspersed with niche attractions, off the beaten path.
198+
199+
The Skill file built by MemOS included:
200+
201+
* Planning afternoon-to-evening, relaxed itineraries.
202+
* Recommending routes that aren't too far or rushed.
203+
* Interspersing niche attractions.
204+
205+
```markdown
206+
# Below is the Skill.md generated by MemOS
207+
---
208+
name: Travel Itinerary Planning
209+
description: Help users plan travel itineraries, ensuring comfortable and efficient exploration of the destination.
210+
211+
---
212+
213+
## Procedure
214+
215+
1. Determine travel purpose and preferences 2. Gather information on destination attractions and activities 3. Filter attractions based on user preferences 4. Arrange daily schedules, including transportation and dining 5. Provide tips and precautions
216+
217+
## Experience
218+
219+
1. Avoid long-distance travel; choose attractions with convenient transportation
220+
2. Reasonably arrange daily schedules, balancing leisure and exploration
221+
3. Fully utilize nighttime activities and attractions to enhance the travel experience
222+
4. Discover niche attractions to avoid crowds and enjoy unique experiences
223+
224+
## User Preferences
225+
226+
- User prefers waking up late and avoiding long-distance travel
227+
- Priority for attractions directly accessible by subway
228+
- Values nighttime activities and experiences
229+
- Explores niche and non-traditional travel routes
230+
231+
## Examples
232+
233+
### Example 1
234+
235+
### Day 1: Giant Panda Afternoon + Niche Old Street Night Tour
236+
- **Noon**: Wake up naturally + Huixinglou Street Food
237+
- **Afternoon**: Chengdu Research Base of Giant Panda Breeding
238+
- **Evening**: Night tour of Shizi Alley + Guihua Alley + Paotongshu Street
239+
...
240+
241+
### Example 2
242+
243+
### Day 2: Three Kingdoms Culture + Dongmen Market Night Market
244+
- **Noon**: Wake up naturally + Wuhouci Street Food
245+
- **Afternoon**: Wuhou Shrine + Red Walls and Bamboo Shadows
246+
- **Evening**: Dongmen Market + Jiuyanqiao Bar Street
247+
...
248+
249+
```
250+
251+
::note
252+
**Start exploring MemOS Skills now! 🚀**
253+
254+
* Go to the [Console - Skill Page](https://memos-dashboard.openmem.net/skill/) to view Skill files automatically generated based on user conversation history.
255+
* Don't have any skills yet? Just [Add Messages](/memos_cloud/mem_operations/add_message) to trigger generation.
256+
::

‎content/en/settings.yml‎

Lines changed: 1 addition & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -30,6 +30,7 @@ nav:
3030
- "(ri:image-add-line) Multimodal Messages": memos_cloud/features/basic/multimodal.md
3131
- "(ri:price-tag-3-line) Custom Tags": memos_cloud/features/basic/custom_tags.md
3232
- "(ri:vip-diamond-line) Advanced Features":
33+
- "(ri:function-line) Skill": memos_cloud/features/advanced/skill.md
3334
- "(ri:book-read-line) Knowledge Base": memos_cloud/features/advanced/knowledge_base.md
3435
- "(ri:tools-line) Tool Calling": memos_cloud/features/advanced/tool_calling.md
3536
- "(ri:chat-history-line) Continuous Dialogue": memos_cloud/features/advanced/continuous_dialogue.md

0 commit comments

Comments
 (0)