dsh-mnemon: Layered, Composable Long-Term Memory for DeepSeek Harness

What Does It Do?

Mnemon is a DSH plugin for DeepSeek Harness that gives the agent layered, visible and composable memory instead of a single flat store. It solves the problem of an agent that starts every session from scratch and keeps asking what it was already told, while avoiding a memory store that either floods each turn or misses what matters. Maintained by omdsh-dev under the MIT license and written in TypeScript, it keeps data local by default and can also connect to third-party memory providers.

Core Features

• Three kinds of memory: runtime memory is injected every turn, project documents are searched first and read when relevant, and memory spaces are recalled on demand from enabled providers.
• Turn memory bar: under each reply it lists the documents and memories that turn read and wrote, and one click opens each in the Memory System.
• One-click save: the brain mark under a reply hands a candidate fact to a task agent that removes duplicates, distils it and writes it to the right memory space, with a receipt showing where it went.
• Compose on the Plugins page: pick a main strategy plus optional enhancements and switch them without moving any data; every component has its own settings page.
• Flexible storage: local by default, or one of several third-party providers, with global, per-workspace or centralized storage and ZIP backup.
• Extensible: sources and strategies are ordinary DSH plugins built on public SDKs, so installed components get the same pages and switches as the shipped ones.

How to Use This Plugin?

After enabling it in DSH, open the Memory System from the sidebar to see status, runtime memory, Project Documents and Memory Spaces; the graph links memories through the entities they mention, and Ask Agent answers with the memories it cites by content. In daily use, when a conversation turns up a fact worth keeping, select the brain mark under the reply, trim the candidate to the sentence worth keeping and send it to the task agent, then use the receipt to jump to the memory space it was written to. To change behaviour, switch the main strategy or add and remove enhancements on the Plugins page, where storage, backup and interface settings also live; the UI guide explains each step.

How to Troubleshoot This Plugin?

If a memory does not show up as expected, first check that the relevant component is enabled on the Plugins page, since strategies and sources take effect through their switches. If a memory landed in the wrong place, edit it directly in the Memory System or verify that the enabled memory space matches the intended provider. For storage and backup issues, review the storage scope and backup configuration in the storage settings as described in the official documentation.