memlikehuman: A Local Memory DSH plugin for DeepSeek Harness Persona and Unfinished Work
xeonphi1/memlikehuman
Gives DeepSeek Harness a local memory: a Python + SQLite MCP service that stores persona, shared history and unfinished work so any model can pick up the story.
memlikehuman is a DSH plugin that keeps persona, shared experiences, and pending tasks on your own machine, written in Python with data stored in SQLite. It targets the pain of losing context when switching models, so old agreements and unfinished work carry into the next conversation. The plugin offers versioned persona rollback, source-tracked facts and preferences, unfinished-work handoff, and budget-aware restore packs; with no API configured, raw persistence, Chinese keyword search, and manual distillation all work offline, alongside import/export and backup.
How to Install
dsh plugin --profile web add github:xeonphi1/memlikehuman- Category
- Memory & Context
- Platform
- DSH Plugin
- Author
- xeonphi1
- Distribution
- Plugin
memlikehuman Key Features
memlikehuman Repo Summary
What Does It Do?
MemLikeHuman is a DSH plugin for DeepSeek Harness that keeps persona, shared experiences, and unfinished work in a local SQLite store and brings them into the next conversation on demand. It solves the problem of losing context when you switch models, so old agreements and pending tasks carry over instead of starting from scratch. Its core capabilities cover versioned persona, layered recall of experiences, source-tracked facts and preferences, work handoff, and budget-aware memory restoration. The project is maintained by xeonphi1 under the Apache-2.0 license and is written in Python.
Core Features
- Persistent persona and experiences: the core persona is stable and versioned with rollback, recent details stay sharp while older events fade into overviews, and raw records remain readable.
- Source-tracked facts and preferences: every distilled memory links back to its origin, distinguishing user statements, assistant claims, and guesses, with confirmation level and change history preserved.
- Unfinished-work handoff: restores goals, completed items, constraints, todos, and next steps so a new model can continue the task.
- Layered forgetting with on-demand readback: older experiences become overviews after 180 days by default and get coarser time labels after 730 days, while exact wording stays available in the original events.
- Automatic capture and pre-compaction handoff: records visible text from a chosen preset, blocks compaction when writes fail, and resumes incrementally after a restart.
- Local-first and controllable data: without any API configured, raw persistence, Chinese keyword search, manual distillation, and restoration all work offline, with import/export and backup support.
How to Use This Plugin?
After enabling it in DSH, initialize the persona and private data directory, then open a new session and confirm the memory tools appear. Keep the same persona_id and project_id across conversations: restore the memory pack first, record new content next, and read the diary when you need exact wording. Day-to-day lookup works by keyword or photo cue, for example searching for a blue kite with a chosen detail level; as memory grows, fast mode uses the local index and assembles a restore pack within your context budget. Long chat histories can be backfilled by exporting, converting, validating, distilling in batches, and checking persona and todos.
This page is an independent rewrite of the plugin's official README — for authoritative documentation and the latest changes, refer to the source: xeonphi1/memlikehuman. The plugin is third-party code that runs on your machine once installed; inclusion does not imply endorsement — please review the source before installing.
DSH Plugin FAQ
