| Repository name |
memscribe |
| Initial maintainer / repository member |
@OLDyade |
| Direction |
Model & Agent System -> Tools & Utilities |
| One-line positioning |
MemScribe is a file-native long-term memory layer for AI Agents, and a memory foundation component inside an Agent Harness. It turns user preferences, project conventions, workflow experience, and contextual facts from long-running tasks into readable, auditable, and portable Markdown files. |
| Repository description |
MemScribe provides SDK, CLI, MCP Server, and Adapter interfaces so different Agent Runtime / Harness systems can standardize access to long-term memory. It does not replace models or Agent Frameworks. Instead, it works as a memory module inside the Harness: joining context assembly during prompt build, capturing new memory after turns, and consolidating or archiving memory during idle phases. |
| Project features |
MemScribe turns memory from a black-box service into governable engineering assets. It uses Markdown files as the source of truth, making memory readable, auditable, and portable. It uses full-index recall instead of opaque top-k retrieval, allowing the main model to keep contextual judgment. It uses tool-calling subagents to extract, update, archive, and consolidate memory, so memory can naturally grow with the Harness lifecycle. The design is lightweight and restrained: zero runtime dependencies, MCP-ready for quick integration, and Adapter-ready for deeper embedding into different Agent Runtime systems. |
| Compliance boundary |
The project does not include model weights, training data, secrets, private credentials, internal configurations, business-specific logic, or proprietary third-party integrations. |
| Expected outcome |
Create a lightweight, open, and reusable Agent Memory tool repository that helps Agent / Harness projects quickly adopt long-term memory, reduce duplicated memory-module development, and accumulate practical engineering experience in Agent systems, Context Engineering, and Harness infrastructure. |
memscribe@OLDyadeModel & Agent System -> Tools & UtilitiesMemScribeis a file-native long-term memory layer for AI Agents, and a memory foundation component inside an Agent Harness. It turns user preferences, project conventions, workflow experience, and contextual facts from long-running tasks into readable, auditable, and portable Markdown files.MemScribeprovides SDK, CLI, MCP Server, and Adapter interfaces so different Agent Runtime / Harness systems can standardize access to long-term memory. It does not replace models or Agent Frameworks. Instead, it works as a memory module inside the Harness: joining context assembly during prompt build, capturing new memory after turns, and consolidating or archiving memory during idle phases.MemScribeturns memory from a black-box service into governable engineering assets. It uses Markdown files as the source of truth, making memory readable, auditable, and portable. It uses full-index recall instead of opaque top-k retrieval, allowing the main model to keep contextual judgment. It uses tool-calling subagents to extract, update, archive, and consolidate memory, so memory can naturally grow with the Harness lifecycle. The design is lightweight and restrained: zero runtime dependencies, MCP-ready for quick integration, and Adapter-ready for deeper embedding into different Agent Runtime systems.