[Feature] Per-User Memory for Agents — Enable Personalized User Understanding Across Conversations #29
Description
Tags: feature-request, ux, api
Quality Rating: ⭐ 8/10
Reporter: xiaoan
Description
Currently, Agents in Clawith do not maintain individual memory for each end-user (C-side user) they interact with. This means every conversation starts from scratch — the Agent has no recollection of a user's preferences, history, or context from previous interactions.
The proposal is to introduce a Per-User Memory system, where each user who interacts with an Agent gets their own dedicated memory store. This would allow the Agent to truly "understand" and remember each individual user over time, creating a much more personalized and human-like experience.
Motivation
- Personalization: Users shouldn't have to repeat their preferences, background, or habits every time. The Agent should remember and proactively apply this knowledge.
- Trust & Engagement: The feeling of being "understood" significantly increases user stickiness and satisfaction — like having a friend who knows you well.
- Efficiency: Reduces repetitive communication. The Agent can make more accurate recommendations and responses based on historical memory.
- Competitive Advantage: Most Agent platforms today are still "stateless" in terms of user understanding. A well-implemented per-user memory system would be a strong differentiator.
Proposed Design Considerations
Memory Structure (Layered Approach)
- Factual Memory — User's basic information and preferences (e.g., "prefers concise answers", "works in finance")
- Interaction Memory — Key conclusions and decisions from past conversations
- Behavioral Memory — User's communication style and tone preferences
Memory Management
- Users should be able to view, edit, and delete their own memory (privacy control)
- Memory should persist across sessions — not disappear when a conversation ends
- A memory decay/relevance strategy should be considered — determining what to keep vs. what to forget over time
Privacy & Compliance
- User data storage must comply with privacy regulations
- Users should have the right to delete all their stored memory
- Clear consent mechanisms should be in place
Accuracy & Correction
- Agent-extracted memories may sometimes be inaccurate — a correction mechanism is needed
- Users should be able to flag or fix incorrect memories
Use Cases
- Customer Service Agent: Remembers a user's past issues, product preferences, and communication style — no need to re-explain context every time.
- Personal Assistant Agent: Knows the user's schedule habits, preferred tools, and working style — provides proactive and tailored assistance.
- Sales/Marketing Agent: Remembers customer interests, past interactions, and buying signals — enables more relevant follow-ups.
Additional Context
This feature would make every user feel like they have a dedicated "person" who truly understands them. When the tools and systems they use can accurately remember them, the overall experience becomes significantly better. This is a highly requested capability in the era of personalized AI interactions.
Submitted via Clawiee (Clawith digital employee) on behalf of xiaoan