memory

Maintain persistent tiered cognitive memory with relevance-scored retrieval and token budgets.

Updated Mar 11, 2026

Search Tags:#knowledge graph #agent orchestration #memory #consolidation #MCP server #token budgeting #tiered recall

Ready to run

npx skills add https://github.com/selfagency/agentsy --skill memory-selfagency

💬 Prompt for ChatGPT / Claude Chat▼

Please help me install this Agent Skill.
Skill: memory
Source: https://github.com/selfagency/agentsy/tree/main/packages/memory/skill
Command: npx skills add https://github.com/selfagency/agentsy --skill memory-selfagency

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve?

Long-running agents lose important context, waste context-window tokens, and struggle to consistently remember preferences, facts, and tool outcomes across sessions.

Core Features & Use Cases

Persistent tiered cognitive memory: Ingest events into sensory-to-long-term tiers with promotion, decay/demotion, and token budget enforcement.
Cross-tier recall and retrieval: Query memories by relevance (including cross-tier search) and list/search within specific tiers for targeted grounding.
Consolidation via awaken cycles: Consolidate queued/pending events, apply decay, and maintain healthier long-horizon knowledge (including persona attributes and a lightweight knowledge graph).

Quick Start

Run the memory MCP server to initialize and operate persistent agent memory for your workflow.

Frequently Asked Questions about memory

High-intent search queries and answers about installing and using this skill.

FAQPage Schema

How do I persist agent memory across sessions so context isn't lost?▼

Persistent agent memory is maintained by ingesting events into sensory-to-long-term cognitive tiers with promotion, decay, and token budget enforcement, enabling context retention and recall across sessions.

What's the best way to manage token budgeting for agent memory retrieval?▼

Agent memory token budgeting is managed through tiered recall with pending queues and awaken cycles, which consolidate queued events, apply decay, and enforce token limits to maintain healthier long-horizon knowledge without overflowing the context window.

How does tiered cognitive memory recall work for AI agents?▼

Tiered cognitive memory recall works by allowing agents to query memories by relevance across sensory-to-long-term tiers, including cross-tier search and targeted listing within specific tiers, applying token budgets to manage retrieval scope.

Can I use an MCP server to maintain a knowledge graph for agent orchestration?▼

Yes, the memory MCP server initializes persistent cognitive memory that includes lightweight knowledge-graph queries and persona storage, supporting agent orchestration by capturing episodic events from tool calls and responses for contextual grounding.

Why does my agent lose important context and waste context-window tokens over time?▼

Agents lose context and waste tokens because they lack persistent memory consolidation; applying tiered cognitive memory with awaken cycles, decay, and relevance-scored retrieval prevents context loss and optimizes token usage across long-running sessions.

When do I need tiered memory consolidation for long-running agents?▼

Tiered memory consolidation is needed when long-running agents must retain preferences, facts, and tool outcomes across sessions, requiring awaken/sleep lifecycle management and decay to maintain relevant long-horizon knowledge within token budgets.