Sapience Memory System & Architecture
Details on Sapience’s memory architecture.
Sapience Memory Overview
Overview
The Agent Memories feature introduces system-level memory capabilities to all Gen3 agents across the Sapience platform. This means your agents can now store, retrieve, and reason over persistent memories—facts, preferences, events, or instructions—across conversations and even across agents. Whether you’re chatting with your personal assistant agent or switching to a specialized feedback or research agent, all agents can now access the same shared knowledge context.
This enhancement significantly improves continuity, personalization, and cross-agent collaboration, making your interactions more intelligent and seamless.
Key Capabilities
- Cross-Agent Memory Recall: Any Gen3 agent can access system-level memories stored by other agents, creating a unified memory layer across the platform.
- Persistent Knowledge Graph: Agents remember structured facts and entities about the user, including preferences, background info, and conversation history context.
- Multi-Modal Memory Triggers: Users can issue natural-language commands like “remember that,” “what do you know about me,” or “list your memories” to store or retrieve data.
- 4-Tier Memory Architecture: The platform introduces four memory layers—custom instructions, chat-level recall, system-wide agentic memory, and ephemeral memory (per session).
- Contextual Lookup: Agents automatically use memory to refine responses, anticipate needs, and follow through on multi-turn or multi-agent workflows.
- Conversation Memory: for Agents with the Memory feature, they can search through your entire Conversation history to look for previous content that’s relevant. Trigger this by telling the Agent that you talked to it before about XYZ and asking it to find that Conversation.
- Explicit Management: You can see the Memories that Sapience is trackng for you at any time by going to Menu > Settings & Customization > Advanced > Manage Memories.
How It Works
Step | Action | Description |
1 | Invoke memory creation | Say things like “remember that my birthday is September 25, 1992” to trigger a memory entry |
2 | Memory stored | The system saves this to your Knowledge Graph as a persistent memory |
3 | Access from any Gen3 agent | You can later ask any agent (even a different one) “what memories do you have about me?” |
4 | Memory response | The agent will list all saved memories and the category (e.g., personal data, preferences) |
5 | Update or remove | You may edit or delete specific memories through support or agent instructions (UI-based tools coming soon) |
Example Use Cases
- User Profile & Personalization
You tell an agent “My timezone is GMT+8” — later, another agent schedules meetings in that timezone automatically.
- Enterprise Knowledge
Your Research Agent remembers “The client prefers quarterly reports” — and this is surfaced automatically during project summaries.
- Cross-Agent Collaboration
You tell your Support Agent “Remember that John is the sales lead for Project X” — and later, your Feedback Agent uses this info during evaluations.
- Task Continuity
Start a file-related conversation with Agent A, then continue it with Agent B days later, without repeating the context.
Tips for Best Results
- Use natural phrasing like: “remember this,” “store that,” or “list what you know about X.”
- Be specific: “Remember that my Azure access expires in March 2026” works better than vague entries.
- Check what’s stored using: “What do you remember about me?”
- Use the memory system to speed up project onboarding, knowledge transfers, and cross-agent tasks.
- Avoid storing confidential or sensitive data unless using enterprise-level encryption tiers.
Screenshots
