Introduction
If you've ever found yourself reintroducing yourself to an AI assistant every time you start a new conversation, or having to re-explain your preferences when switching projects, this tutorial is for you. WorkBuddy's 3-layer memory system is designed to solve these exact problems, allowing AI to truly "remember" you and adapt to your workflow over time.
What is the 3-Layer Memory System?
WorkBuddy's memory system is structured into three layers, each serving a unique purpose and working together to create a seamless, personalized experience:
- Layer 1: Cloud Memory
- Automatically learns your habits and preferences over time
- Generates a personal profile that works across all conversations
- Requires no manual management
- Layer 2: User-level Local Memory
- Stores information you explicitly ask WorkBuddy to remember
- Example: "Use this design style for this series of projects"
- Saved in a local file
- Layer 3: Workspace Memory
- Maintains independent records for each project
- Automatically adds daily logs after completing substantive work
- Requires no manual effort from you
These three layers work together, moving from broad to specific, from passive to active, to create a comprehensive memory system.
Deep Dive into Each Layer
Layer 1: Cloud Memory
Cloud memory has two core capabilities:
1. Auto Profile Generation
- Servers analyze your conversations to automatically summarize your code preferences, work habits, and commonly used tools
- This profile is automatically loaded every time you start a new conversation
- WorkBuddy "recognizes" you without you having to reintroduce yourself
2. Cross-Conversation History Search
- When you mention something like "the deployment solution we discussed last week," WorkBuddy uses
conversation_searchto search through all your historical conversations - It can find the exact discussion, including tools used, configurations, and ports
- You don't have to re-explain everything — it finds the information itself
Layer 2 & 3: Local Memory
Layer 2: User-level Local Memory
- Stored in the
MEMORY.mdfile - When you explicitly state your requirements, those rules are written to this file
- Applies to all projects and conversations moving forward
Layer 3: Workspace Memory
- Stored in the
.workbuddy/memory/directory within your project folder - After you complete substantive work, WorkBuddy automatically adds a daily log entry
- Logs include what you did, what solutions you used, and what problems you encountered
- Logs older than 30 days are refined into long-term project notes
Memory System + Skills + Automation: The Complete Workflow
Skills System
- Memory is the "brain" of WorkBuddy, while the skills system is its "hands"
- Skills are divided into three categories: user-level, project-level, and built-in marketplace
- A skill is a reusable workflow
- Example: If you have the
poster-design-warm-techskill installed, simply say "help me create a tech poster" and WorkBuddy automatically loads the skill's context and generates it according to the preset workflow - After completing a complex workflow, WorkBuddy can automatically accumulate it as a new skill, getting smarter with use
Automation System
- Automation is the most "active" form of the memory system
- Supports both recurring tasks and one-time tasks
- Examples:
- "Generate a market briefing every morning at 9 AM"
- "Write a weekly report automatically every Friday at 5 PM"
- Can specify which expert to activate automatically and which third-party service to connect to
In simple terms:
- Cloud memory is passively learned
- Local memory is static rules
- Workspace memory is post-work records
- Automation is "pre-set future memory" that automatically wakes up and executes at specified times
Practical Operation Guide
Setting Up Automation Tasks
- Open WorkBuddy
- Click "Automation" → "Add Task"
- Fill in: Name, Workspace, Prompt, Connector, Execution frequency
Each configuration item corresponds to a piece of memory: the prompt is the "memory body" of automation, and the expert and connector determine the context scope.
Key Settings Interfaces
1. Memory Settings
The management center for cloud memory. Here you can toggle automatic profile generation, manage existing memory cards, and import habits from other AI platforms.
2. Conversation Management
The cost switch for cloud memory. Automatically disconnects context after long periods of inactivity to save tokens. With conversation_search, disconnecting doesn't mean forgetting.
3. Personalization
The rule engine for local memory. Custom instructions written here are saved to the MEMORY.md file. Applies to all projects with the highest priority.
4. Data Management
The archive for workspace memory. Centralized display of shared files, archived tasks, and installed applications for all projects.
5. Workspace
The write trigger for workspace memory. Manually save content to .workbuddy/. Works with automatic log addition to form a complete memory loop.
Summary
Good AI assistants aren't one-time tools — they remember your every decision and help you accumulate every experience.
- Use Memory Wisely: Say "remember this" to save rules locally. Say "the solution we discussed last time" to use cloud memory. Let it automatically write logs after completing important work.
- Accumulate Skills: Complex workflows are automatically accumulated. Bug fixes are automatically learned. Reusable workflows mean you don't have to start from scratch every time.
- Free Your Hands: Set up automation to handle recurring tasks. Let WorkBuddy work for you even when you're not around.
This is the WorkBuddy memory system — making AI understand you better, work more efficiently, and become a true partner in your workflow.
常见问题
How is WorkBuddy's memory different from ChatGPT's memory?
ChatGPT's memory is a single flat layer — it remembers facts you tell it, but doesn't distinguish between personal preferences, project-specific context, and daily work logs. WorkBuddy's 3-layer system separates these concerns: Cloud Memory handles cross-conversation personalization (like "I prefer blue themes"), Local Memory stores explicit rules (like "always use this brand voice"), and Workspace Memory auto-logs what you actually did each day. The key difference: WorkBuddy doesn't just remember facts — it remembers workflows. After you complete a complex task, it can automatically save the sequence as a reusable Skill, so next time you just say "do the weekly report" instead of re-explaining the entire process.
Is my cloud memory data private?
Cloud memory is processed on Tencent's servers, so it's subject to Tencent's privacy policy. The auto-profile generation and cross-conversation search require server-side processing. If you're concerned about sensitive data, use Local Memory (MEMORY.md) and Workspace Memory (.workbuddy/memory/) — both are stored entirely on your machine and never leave it. You can also toggle off automatic profile generation in Memory Settings while keeping manual cloud memory features. For maximum privacy, use WorkBuddy in local-only mode with cloud features disabled.
How long does it take for Cloud Memory to build a useful profile?
About 1-2 weeks of regular use (10-20 conversations). The profile starts with surface-level preferences (preferred tools, common tasks) and deepens over time. After a month, it should reliably know your coding style, design preferences, commonly used project structures, and even your communication tone. The key is consistency — using WorkBuddy as your primary AI assistant accelerates profile building. You can also import habits from other AI platforms in Memory Settings to jumpstart the process.
Can I use Automation without setting up Memory first?
Yes, but you'll get less value. Automation without memory is just a scheduler — it runs the same prompt at the same time every day. With memory, Automation becomes contextual: the weekly report knows what you worked on this week (from Workspace Memory), uses your preferred format (from Local Memory), and adapts to your communication style (from Cloud Memory). Start with simple time-based automations, and as your memory layers fill in, the same automations become progressively smarter without any additional configuration.