[Plugin Request]: Long-term Memory Assistant - az_agent_long_memory #503

New Issue

2026-02-22T17:30:50-05:00

yindo commented

2026-02-22 17:30:50 -05:00

Originally created by @yuntaosong on GitHub (Feb 8, 2026).

Plugin Name

az_agent_long_memory

Function Description

Plugin Feature Description

AZMemory Long-term Memory Plugin for Intelligent Assistant

I. Functional Overview
This plugin empowers the intelligent assistant with core long-term memory capabilities. By automatically storing and managing historical dialogue data between users and the assistant, it precisely retrieves user preferences, behavioral habits, and background information relevant to the current interaction. This enables the assistant to achieve "continuous recognition and understanding" of the user. Users no longer need to repeatedly explain basic circumstances, as every conversation can provide more tailored responses based on historical accumulation, significantly enhancing the coherence and intelligence of the interaction.

II. Core Modules

(1) Memory Retrieval Module

Functional Positioning
Based on the user's current query intent, quickly match and extract relevant memory fragments from the historical dialogue to provide accurate contextual support for the Large Language Model (LLM).
Key Parameters
- query: Required. The user's current query content, used for semantic matching of relevant historical memories.
- user_id: System field. The unique identifier for the user.
Output Format
Returns JSON structured data, primarily containing the memory field:
- Relevant Memory Exists: Returns the matched historical memory fragment (integrated natural language text).
- No Relevant Memory: Returns the fixed prompt "No memory found".
Applicable Scenarios
Triggered before the LLM generates a response, injecting the retrieved memory information into the prompt. This allows the model to generate personalized responses based on the user's historical interactions. (For example, if a user previously mentioned "working at night," the model can recommend 24-hour delivery services when consulting about coffee purchases later).

(2) Memory Update Module

Functional Positioning
Real-time saving of the complete dialogue records between the user and the assistant (including user queries and the assistant's final responses), continuously enriching the long-term memory database and providing a stable, comprehensive data foundation for memory retrieval.
Key Parameters
- query: Required. The user's current query content, used for subsequent semantic matching association.
- agent_response: Required. The final response presented to the user after the assistant completes all calculations.
- user_id: System field. The unique identifier for the user, configured consistently with the "Memory Retrieval Module".
Applicable Scenarios
Automatically triggered at the end of each dialogue to upload the complete interaction pair of "user query + assistant response" to the memory database, achieving real-time updating and accumulation of long-term memory. (For example, when a user adds new preferences or supplements background information, it can be instantly stored in the memory database for subsequent retrieval).

III. Core Value

Personalized Interaction Upgrade: Precisely captures long-term user preferences (e.g., "likes dark roast coffee"), behavioral habits (e.g., "works at night"), and background information, making the assistant's responses more understanding and warmer.
Efficient Interaction Experience: Completely avoids users having to repeatedly explain basic information (e.g., occupation, demand preferences), reducing ineffective communication costs and improving dialogue efficiency.
Intelligent Capability Accumulation: As the number of interactions increases, the memory database continues to enrich. The model's recommendations and response logic become more accurate and tailored to the user's personalized needs, achieving an "smarter with use" intelligent evolution.

IV. Application Scenarios

Intelligent E-commerce Service Assistant: When a user consults about product selection, call the memory_search tool to retrieve their past shopping preferences (e.g., "likes cost-effective domestic brands," "needs products suitable for sensitive skin") and feedback related to historical orders (e.g., "previously bought XX model appliance, concerned about after-sales guarantee"). Based on this memory, recommend suitable products without repeatedly asking for basic information like skin type, budget, or brand preference. After the conversation, call the memory_update tool to save the product needs and final purchase decision from this consultation, improving subsequent recommendation dimensions.
Intelligent Customer Service Assistant: When a user consults about communication package issues, call the memory_search tool to extract their historical package usage habits (e.g., "consumes about 20GB of data per month, frequently uses voice calls") and past consultation records (e.g., "raised concerns about package deductions last month"). Directly recommend a suitable package upgrade plan based on this memory and address注意事项 related to historical inquiries, avoiding the need to reconfirm "how much data is used monthly" or "if voice call functionality is needed."
Intelligent Life Assistant: When a user asks for "weekend getaway recommendations," call the memory_search tool to retrieve their travel preferences (e.g., "likes natural landscapes, dislikes crowded scenic spots," "needs convenient parking") and feedback from past trips (e.g., "liked the hiking trail at XX mountain previously"). Recommend niche scenic spots and travel plans based on this memory. After the conversation, use the memory_update tool to save the travel needs from this session, accumulating data for future holiday recommendations.
Intelligent Learning Partner: When a user seeks help with English learning, call the memory_search tool to obtain their past learning progress (e.g., "preparing for CET-4, weak in listening") and learning habits (e.g., "likes to practice listening at 8 PM every night"). Recommend suitable listening materials and practice methods accordingly, avoiding the repetition of explaining already mastered vocabulary or grammar points.
Intelligent Workplace Assistant: When a user consults about "writing a work summary," call the memory_search tool to extract their past summary preferences (e.g., "accustomed to a three-part structure of 'Work Achievements - Problem Improvement - Next Month's Plan'," "needs to highlight data support") and industry characteristics (e.g., "Internet operations position, focusing on user growth data"). Provide a writing framework and key directions based on this memory, avoiding the need to reconfirm "what structure do you usually use for your summaries" or "what core content needs to be highlighted."
Intelligent Maternal and Infant Service Assistant: When a user consults about childcare issues, call the memory_search tool to retrieve the baby's age (e.g., "the baby is 10 months old, currently in the complementary food addition stage"), past childcare preferences (e.g., "inclined towards scientific feeding, chooses organic complementary food"), and previously concerned pain points (e.g., "the baby doesn't like to eat vegetables"). Provide suitable complementary food recipes and feeding techniques based on this memory, avoiding the need to repeatedly ask for basic information like the baby's age or feeding method.
Intelligent Fitness Guidance Assistant: When a user consults about a fitness plan, call the memory_search tool to extract their past fitness foundation (e.g., "a beginner, previously tried jogging"), physical condition requirements (e.g., "an office worker who sits for long periods, needs actions to relieve waist fatigue"), and fitness goals (e.g., "lose 5 catties in 3 months"). Recommend a home workout plan based on this memory, avoiding the need to reconfirm information like fitness foundation, goals, or body parts with discomfort.

Official Website URL

https://azmem.tech/

Originally created by @yuntaosong on GitHub (Feb 8, 2026). ### Plugin Name az_agent_long_memory ### Function Description Plugin Feature Description AZMemory Long-term Memory Plugin for Intelligent Assistant I. Functional Overview This plugin empowers the intelligent assistant with core long-term memory capabilities. By automatically storing and managing historical dialogue data between users and the assistant, it precisely retrieves user preferences, behavioral habits, and background information relevant to the current interaction. This enables the assistant to achieve "continuous recognition and understanding" of the user. Users no longer need to repeatedly explain basic circumstances, as every conversation can provide more tailored responses based on historical accumulation, significantly enhancing the coherence and intelligence of the interaction. II. Core Modules (1) Memory Retrieval Module - Functional Positioning Based on the user's current query intent, quickly match and extract relevant memory fragments from the historical dialogue to provide accurate contextual support for the Large Language Model (LLM). - Key Parameters - query: Required. The user's current query content, used for semantic matching of relevant historical memories. - user_id: System field. The unique identifier for the user. - Output Format Returns JSON structured data, primarily containing the memory field: - Relevant Memory Exists: Returns the matched historical memory fragment (integrated natural language text). - No Relevant Memory: Returns the fixed prompt "No memory found". - Applicable Scenarios Triggered before the LLM generates a response, injecting the retrieved memory information into the prompt. This allows the model to generate personalized responses based on the user's historical interactions. (For example, if a user previously mentioned "working at night," the model can recommend 24-hour delivery services when consulting about coffee purchases later). (2) Memory Update Module - Functional Positioning Real-time saving of the complete dialogue records between the user and the assistant (including user queries and the assistant's final responses), continuously enriching the long-term memory database and providing a stable, comprehensive data foundation for memory retrieval. - Key Parameters - query: Required. The user's current query content, used for subsequent semantic matching association. - agent_response: Required. The final response presented to the user after the assistant completes all calculations. - user_id: System field. The unique identifier for the user, configured consistently with the "Memory Retrieval Module". - Applicable Scenarios Automatically triggered at the end of each dialogue to upload the complete interaction pair of "user query + assistant response" to the memory database, achieving real-time updating and accumulation of long-term memory. (For example, when a user adds new preferences or supplements background information, it can be instantly stored in the memory database for subsequent retrieval). III. Core Value - Personalized Interaction Upgrade: Precisely captures long-term user preferences (e.g., "likes dark roast coffee"), behavioral habits (e.g., "works at night"), and background information, making the assistant's responses more understanding and warmer. - Efficient Interaction Experience: Completely avoids users having to repeatedly explain basic information (e.g., occupation, demand preferences), reducing ineffective communication costs and improving dialogue efficiency. - Intelligent Capability Accumulation: As the number of interactions increases, the memory database continues to enrich. The model's recommendations and response logic become more accurate and tailored to the user's personalized needs, achieving an "smarter with use" intelligent evolution. IV. Application Scenarios - Intelligent E-commerce Service Assistant: When a user consults about product selection, call the memory_search tool to retrieve their past shopping preferences (e.g., "likes cost-effective domestic brands," "needs products suitable for sensitive skin") and feedback related to historical orders (e.g., "previously bought XX model appliance, concerned about after-sales guarantee"). Based on this memory, recommend suitable products without repeatedly asking for basic information like skin type, budget, or brand preference. After the conversation, call the memory_update tool to save the product needs and final purchase decision from this consultation, improving subsequent recommendation dimensions. - Intelligent Customer Service Assistant: When a user consults about communication package issues, call the memory_search tool to extract their historical package usage habits (e.g., "consumes about 20GB of data per month, frequently uses voice calls") and past consultation records (e.g., "raised concerns about package deductions last month"). Directly recommend a suitable package upgrade plan based on this memory and address注意事项 related to historical inquiries, avoiding the need to reconfirm "how much data is used monthly" or "if voice call functionality is needed." - Intelligent Life Assistant: When a user asks for "weekend getaway recommendations," call the memory_search tool to retrieve their travel preferences (e.g., "likes natural landscapes, dislikes crowded scenic spots," "needs convenient parking") and feedback from past trips (e.g., "liked the hiking trail at XX mountain previously"). Recommend niche scenic spots and travel plans based on this memory. After the conversation, use the memory_update tool to save the travel needs from this session, accumulating data for future holiday recommendations. - Intelligent Learning Partner: When a user seeks help with English learning, call the memory_search tool to obtain their past learning progress (e.g., "preparing for CET-4, weak in listening") and learning habits (e.g., "likes to practice listening at 8 PM every night"). Recommend suitable listening materials and practice methods accordingly, avoiding the repetition of explaining already mastered vocabulary or grammar points. - Intelligent Workplace Assistant: When a user consults about "writing a work summary," call the memory_search tool to extract their past summary preferences (e.g., "accustomed to a three-part structure of 'Work Achievements - Problem Improvement - Next Month's Plan'," "needs to highlight data support") and industry characteristics (e.g., "Internet operations position, focusing on user growth data"). Provide a writing framework and key directions based on this memory, avoiding the need to reconfirm "what structure do you usually use for your summaries" or "what core content needs to be highlighted." - Intelligent Maternal and Infant Service Assistant: When a user consults about childcare issues, call the memory_search tool to retrieve the baby's age (e.g., "the baby is 10 months old, currently in the complementary food addition stage"), past childcare preferences (e.g., "inclined towards scientific feeding, chooses organic complementary food"), and previously concerned pain points (e.g., "the baby doesn't like to eat vegetables"). Provide suitable complementary food recipes and feeding techniques based on this memory, avoiding the need to repeatedly ask for basic information like the baby's age or feeding method. - Intelligent Fitness Guidance Assistant: When a user consults about a fitness plan, call the memory_search tool to extract their past fitness foundation (e.g., "a beginner, previously tried jogging"), physical condition requirements (e.g., "an office worker who sits for long periods, needs actions to relieve waist fatigue"), and fitness goals (e.g., "lose 5 catties in 3 months"). Recommend a home workout plan based on this memory, avoiding the need to reconfirm information like fitness foundation, goals, or body parts with discomfort. ### Official Website URL https://azmem.tech/

yindo referenced this issue

2026-02-22 17:44:22 -05:00

[PR #503] [MERGED] add knowledge plugin #981

Sign in to join this conversation.

1 Participants

Notifications

Due Date

No due date set.

Dependencies

No dependencies set.

Reference: langgenius/dify-plugins#503