Skip to content
Back to search

dsh-memoknow

Local personal memory and knowledge management for DeepSeek Harness. It supports local FTS, CPU-embedding or OpenAI-Compatible Embedding API.

Safety scan

A static text scan, not a guarantee. Select the badge to see what was scanned and every finding.

Checked today
Badge for your README
findsafeskills safety rating

Markdown

[![findsafeskills safety rating](/api/listings/twoyoung91--dsh-memoknow/badge.svg)](/listings/twoyoung91--dsh-memoknow)

HTML

<a href="/listings/twoyoung91--dsh-memoknow"><img src="/api/listings/twoyoung91--dsh-memoknow/badge.svg" alt="findsafeskills safety rating" /></a>

Install

i

Install instructions from the README

As written by the repo's author. findsafeskills has not verified these steps.

Download the `.tgz` asset from [GitHub Releases](https://github.com/twoyoung91/dsh-memoknow/releases/latest). MemoKnow v0.2.0 supports Harness `0.2.0-rc.1` and the declared compatible 0.1.x tool API versions. Install the package into your web profile:

dsh plugin --profile web add C:\path\to\dsh-external-dsh-memoknow-0.2.0.tgz
dsh web

Restart DSH if it is already running, then open **Settings → MemoKnow**. The release's `.tgz` includes the built plugin; GitHub's automatic source archives require the source-build steps below.

What was scanned

The files the safety scan read, as written by the repository's author and shown as plain text. Lines with a finding are marked.

README.md · README · 222 lines
  1. ![MemoKnow — Local memory and knowledge for DeepSeek Harness](docs/banner.svg)
  2. [English](README.md) | [简体中文](README.zh-CN.md)
  3. MemoKnow is a local DeepSeek Harness plugin for personal memory and
  4. user-imported knowledge. It stores structured records in SQLite, snapshots
  5. original documents by SHA-256, combines FTS5 with optional sqlite-vec semantic
  6. retrieval, exposes agent tools, and provides a management screen inside DSH.
  7. MemoKnow is an independent community plugin, not an official DeepSeek Harness
  8. component. It is designed for one person's local DSH profile; it is not a
  9. multi-user service or a cloud sync product.
  10. ## What it does
  11. - Distills durable facts, preferences, and decisions from eligible completed
  12. chat turns; explicit “please remember” requests are processed sooner.
  13. - Provides a review inbox in DSH **Settings → MemoKnow**: inspect source
  14. sessions and turns, edit suggestions, and approve or dismiss them. Approved
  15. memories stay intact until you accept a proposed replacement.
  16. - Lets you pause automatic learning, exclude individual sessions, adjust token
  17. budgets, and inspect pending work and daily usage.
  18. - Imports text/Markdown, Word, PDF, CSV, and Excel documents as immutable local
  19. knowledge snapshots with searchable text. Search the knowledge library,
  20. preview extracted passages, and inspect or retry semantic indexing.
  21. Embedded images are ignored.
  22. - Lets you edit or permanently forget individual memories. Normal recall
  23. includes only active, non-expired records.
  24. - Starts with Local FTS, so no embedding model or API key is needed. Optional
  25. CPU or OpenAI-compatible embeddings improve knowledge retrieval.
  26. ## A quick look
  27. Open **DSH Settings → MemoKnow** to review memories, explore your knowledge
  28. library, and control what the plugin learns.
  29. Review suggested memories and inspect their evidence before approving them:
  30. ![DSH Settings showing MemoKnow's memory review inbox and an example suggestion](docs/screenshots/memory-library.png)
  31. Search imported knowledge and preview its extracted text:
  32. ![DSH Settings showing MemoKnow's knowledge library and extracted-text preview](docs/screenshots/knowledge-import.png)
  33. Pause learning, set token budgets, and inspect pending sessions:
  34. ![DSH Settings showing MemoKnow's learning controls, usage, and pending demo session](docs/screenshots/learning-controls.png)
  35. Start with Local FTS and adjust retrieval settings when needed:
  36. ![DSH Settings showing MemoKnow retrieval settings with Local FTS selected](docs/screenshots/retrieval-settings.png)
  37. For step-by-step setup, daily use, backups, and troubleshooting, read the
  38. [user manual](docs/USER_MANUAL.md). Contributors should read
  39. [CONTRIBUTING.md](CONTRIBUTING.md); release changes are in
  40. [CHANGELOG.md](CHANGELOG.md).
  41. ## Install the latest release
  42. Download the `.tgz` asset from [GitHub Releases](https://github.com/twoyoung91/dsh-memoknow/releases/latest).
  43. MemoKnow v0.2.0 supports Harness `0.2.0-rc.1` and the declared compatible 0.1.x
  44. tool API versions. Install the package into your web profile:
  45. ```powershell
  46. dsh plugin --profile web add C:\path\to\dsh-external-dsh-memoknow-0.2.0.tgz
  47. dsh web
  48. ```
  49. Restart DSH if it is already running, then open **Settings → MemoKnow**.
  50. The release's `.tgz` includes the built plugin; GitHub's automatic source archives
  51. require the source-build steps below.
  52. ## Quick start from a checkout
  53. Requires Node.js `^22.19.0 || >=24.0.0`, pnpm 11.7, and a compatible DSH
  54. installation. Run these commands from the MemoKnow checkout:
  55. ```powershell
  56. pnpm install --frozen-lockfile
  57. pnpm run check
  58. dsh plugin --profile web add .
  59. ```
  60. If your DSH installation is a source checkout and `dsh` is not on `PATH`, use
  61. `pnpm dsh plugin --profile web add <absolute-path-to-MemoKnow>` from the DSH
  62. repository root instead. To install a prebuilt GitHub Release tarball, use
  63. `dsh plugin --profile web add <path-to-tarball>`. Restart the web profile after
  64. installation. Direct `github:` installation is not supported yet: Git installs
  65. source files, and this package does not currently build during Git install.
  66. These commands follow DSH's [plugin packaging guide](https://github.com/deepseek-ai/deepseek-harness/blob/master/docs/user/develop/basic/publish.md)
  67. and [CLI reference](https://github.com/deepseek-ai/deepseek-harness/blob/master/apps/cli/reference/README.md).
  68. Restart DSH after installing, open **Settings → MemoKnow**, and select
  69. **Setup & settings**. Local FTS is the default and needs no model. Local CPU
  70. embedding downloads the pinned INT8 `Xenova/multilingual-e5-small` model on
  71. first enable (about 140 MiB including tokenizer files). OpenAI-compatible mode
  72. validates its endpoint and model before activation. FTS remains available if an
  73. embedding provider later becomes unavailable.
  74. If API mode is selected, put the key in the named environment variable (default
  75. `MEMOKNOW_EMBEDDING_API_KEY`). MemoKnow stores only that variable's name, never
  76. the secret value.
  77. ## Managed embedding policy
  78. Products that embed MemoKnow can enforce one OpenAI-compatible embedding
  79. provider through the standard Cordis plugin config while keeping the public
  80. plugin source unchanged:
  81. ```yaml
  82. - id: dsh-memoknow
  83. name: '@dsh-external/dsh-memoknow'
  84. config:
  85. embedding:
  86. source: managed-api
  87. baseUrl: https://managed.example/v1
  88. model: managed-embedding-model
  89. apiKeyEnv: MEMOKNOW_EMBEDDING_API_KEY
  90. ```
  91. Managed mode projects these values into the runtime settings and disables their
  92. controls in the management page. Settings API updates cannot override the
  93. policy. The endpoint must use HTTPS, except for loopback development addresses.
  94. Only the environment-variable name belongs in config; the API key value must be
  95. provided through the process environment and is never returned by MemoKnow.
  96. ## Local data
  97. The default data root is `$DSH_HOME/memoknow`, or `~/.dsh/memoknow` when
  98. `DSH_HOME` is unset:
  99. ```text
  100. memoknow.sqlite3
  101. objects/sha256/aa/bb/<full-sha256>[.txt]
  102. models/ # created only after Local CPU is enabled
  103. ```
  104. Use the plugin `dataDir` setting in the Cordis patch to choose another root.
  105. SQLite is authoritative for metadata and lifecycle state. Original document
  106. snapshots are immutable files. FTS and sqlite-vec indexes are derived data.
  107. ## Knowledge imports
  108. The management page accepts pasted text/Markdown and files up to 25 MiB:
  109. - Word `.doc` and `.docx`
  110. - PDF with an extractable text layer
  111. - CSV encoded as UTF-8
  112. - Excel `.xlsx`
  113. The exact original bytes are retained. Embedded images are not extracted,
  114. embedded, or indexed. Scanned image-only PDFs therefore require OCR, which is
  115. not part of this release. Legacy Excel `.xls` is not supported yet.
  116. ## Agent tools
  117. - `memoknow_remember`
  118. - `memoknow_search`
  119. - `memoknow_memory_list`
  120. - `memoknow_memory_update`
  121. - `memoknow_memory_forget`
  122. - `memoknow_knowledge_import`
  123. - `memoknow_knowledge_remove`
  124. Memory age lowers retrieval rank toward a floor but does not delete records.
  125. Forget removes a memory row and its search-index entry without creating a
  126. tombstone. Automatic tombstone maintenance for outdated records is a design
  127. goal, not a feature of this version. Forgetting a derived memory does not delete
  128. its originating DSH chat session or other backups. Knowledge removal deletes the
  129. document/index records; a content-addressed original is deleted only after its
  130. final reference is removed.
  131. The default memory half-life is 180 days and automatic recall returns at most 12
  132. memory records. Knowledge has no user-configured result cap; retrieval still uses
  133. an internal safety ceiling to protect the agent context and SQLite process.
  134. ## Automatic memory updates
  135. After a completed root-agent turn, MemoKnow records only new direct-user and
  136. visible assistant text. It excludes failed turns, subagents, tool/plugin context,
  137. reasoning blocks, trivial acknowledgements, and credential-like content. This
  138. local capture does not call a model and advances a durable per-session checkpoint.
  139. Eligible turns are distilled with the model already configured for that DSH
  140. session. Ordinary updates are batched for two minutes or five eligible turns;
  141. an explicit request such as “please remember” bypasses the delay. The model sees
  142. only the pending delta plus a small set of lexically related memories, has no
  143. tools, is limited to 700 output tokens and a 45-second call, and must return
  144. validated JSON. Inferred memories enter as `candidate`; explicitly requested
  145. memories may enter as `active`. Automatic changes to approved memories become
  146. separate replacement candidates. Dismissal keeps the original; approval replaces
  147. it only if it has not changed since the proposal was created.
  148. Capture and processing are separate transactions. A restart, timeout, model
  149. failure, revision conflict, or token-budget refusal leaves captured turns pending
  150. for a later retry. Successful memory changes, usage accounting, and checkpoint
  151. advancement commit atomically. Defaults cap automatic distillation at 8,000
  152. tokens per session and 30,000 tokens per UTC day.
  153. ## Security notes
  154. The management API is same-origin, rejects cross-site mutations, limits JSON
  155. bodies to 1 MiB and file requests to 26 MiB, validates file signatures, uses
  156. parameterized SQL, sets restrictive browser headers, and never returns secret
  157. settings. Keep the DSH web host bound to loopback unless you add an
  158. authenticated reverse proxy.
  159. Please report security concerns through [SECURITY.md](SECURITY.md) rather
  160. than a public issue.
  161. ## Current limitations
  162. - Automatic distillation uses the active DSH session model; there is not yet a
  163. separate model selector. Token budgets are configurable in **Learning**.
  164. - PDF import does not OCR scanned pages and deliberately ignores images.
  165. - Local CPU embedding runs on CPU through Transformers.js. Initial download and
  166. indexing can take time on slower connections or large libraries.
  167. - Excel `.xls`, password-protected files, macros, and embedded images are not
  168. imported.
  169. - Document previews show extracted text with overlapping passages, not the
  170. original layout or clickable page-level citations. Deletes use confirmation
  171. dialogs; memory edits use a dedicated review form.

Ratings and comments

Sign in to rate this skill.

Sign in

Free to use. If it’s helped, pass it on.