PerryLink/dsh-memento
Bounded, layered, approval-gated, auditable cross-session memory: a typed `ctx.memory` seam with a zero-dependency SQLite provider, a `memory` tool, and frozen snapshot injection; every write passes the approval gate and stays reconstructable from the session log.
Install
via npm
Resolves a published tarball rather than cloning the repository, and installs without any extra setup. Swap `web` for your profile name if you run another one.
via GitHub · npm package
Installing from GitHub runs the project's build script, which pnpm blocks until you allowlist it — run the command once and pnpm prints the exact key to add under `allowBuilds` in ~/.dsh/profiles/web/pnpm-workspace.yaml.
What happened when we ran it
Installed cleanly when we ran it
Every command here is run in a throwaway container against a clean profile, and the result is whatever the harness recorded — not a guess from the source. Last run 2d ago.
Due diligence
What it is — Provides a typed memory seam with approval gate and audit trails for DeepSeek Harness sessions.
Who it is for — Users who need to record user preferences or agent environment norms across sessions are suited. Users who only need to record environment facts within a single session can choose other solutions.
Watch out — Successfully tested in a fresh profile where it registers into the profile via harness. No obvious issues found.
The verdict — I would install it because it has hard budgets per track and layer, causing full stores to fail with structured error and model consolidate retry.
Generated by grok-4.6, and a starting point rather than a verdict. Where it says a plugin installs or does not, that is from a real run in a clean profile — everything else is read off the repository. Trust the source over this.
dsh-memento is a DeepSeek Harness plugin that provides bounded, cross-session memory through a typed ctx.memory service, a local SQLite store, a memory tool and a frozen prompt snapshot. DeepSeek Harness loads the service for each configured agentPreset. The store uses node:sqlite with WAL at $DSH_HOME/dsh-memento/memory.db by default, and separates user and agent tracks into user-global and workspace layers. The snapshot is assembled once at the first prompt and does not change during the session.
Every add, replace, remove or seed write passes the service-level approval policy set by writePolicy or writePolicies; denied writes still produce audit rows. Hard per-track and per-layer character budgets reject full stores rather than truncating or compacting them. The memory_recall tool can also search bounded recent session history, while /memory provides listing, querying, proposals, budgets, audits and import/export.
dsh-memento suits agents that need small, inspectable preferences, environment facts or conventions across sessions without a separate memory service. It is a poor choice for large knowledge bases or retrieval-heavy workloads: search is small-corpus substring matching and storage is deliberately bounded. The plugin reaches the terminal surface through its /memory command and reads or writes the local database and session-log-related audit inputs; the README states that it uses no network and no native code.
DeepSeek Harness reads the plugin settings from the memento row in cordis.yml. Important defaults include:
| Key | Default | Meaning |
|---|---|---|
enabled |
true |
Enables the service, tools, snapshot, command, panel and answerer |
dbPath |
'' |
Uses $DSH_HOME/dsh-memento/memory.db; relative paths resolve under $DSH_HOME |
writePolicy |
ask |
Accepts ask, auto or off |
language |
en |
Model-visible and command-output language; also accepts zh |
maxEntriesPerQuery |
20 |
Query result cap, hard-capped at 1000 |
auditRetentionDays |
0 |
Keeps audit records forever when 0 |
proposals.enabled |
true |
Captures a proposal after successful compaction |
The hard character budgets default to 2000 for both user layers and 4000 for both agent layers. snapshotOrder defaults to -50; commandListLimit and commandAuditLimit default to 50 and 10. Recall defaults are historyLimitDefault: 8, snippetCap: 5, snippetChars: 300 and windowDays: 30. proposals.maxChars is 2000 and proposals.maxPending is 8.
DeepSeek Harness adds /memory with list, query, add, remove, consolidate, proposals, budgets, audit, export, import <path> and adapters. The memory tool supports add, replace, remove, consolidate and query; memory_recall returns bounded memory and recent session-history matches. The web panel is read-only and shows entries, searches, budget bars and the audit tail.
The database uses node:sqlite, WAL and file mode 0600. user and agent tracks each have user-global and workspace layers, isolated per agentPreset. Writes are routed through the approval waterfall in the service. The injected snapshot is written verbatim to request/header.system; approval and audit records support reconstruction of model-visible content.
DeepSeek Harness compatibility is 0.1.0-rc.6; Node must be ^22.19.0 || >=24.0.0. Windows, macOS and Linux are supported. The README describes a pure-host implementation with no native code or network. Full budgets return a structured usage-and-limit error: entries are not truncated or automatically compacted.
Written from the project's own documentation and kept in sync with it. Where the two disagree, the source is authoritative — read the README on GitHub
Same category
volcengine
OpenViking memory and context bundle for DeepSeek Harness: pre-step auto-recall and profile injection, session capture, `viking://` URI guarding, and recall/write memory tools backed by an OpenViking server.
No one-line install — this plugin lives inside a larger repository and publishes no npm package.
vectorize-io
Hindsight, agent memory that learns: long-term project memory with auto recall and retain, knowledge pages, deep reflection, and per-repo memory banks.
No one-line install — this plugin lives inside a larger repository and publishes no npm package.
bowenliang123
Context insight panel: see what the model's context window is made of and how it evolves — composition vs. window size, per-request history, compression/injection events, and per-message token stats.
Installed cleanly when we ran it
csyangwen
为 DeepSeek Harness 带来「跨会话长期记忆 + 后台自我进化」能力的纯插件实现：五轨记忆 · git 分支感知 · 回合内自我审查 · 技能自我进化与技能管理器 · 四轨待办 · COI 调度 · 会话广播 · 会话搜索 · 提示词管理器 · 临时信息便签——零核心修改、零运行时依赖，随装随用、卸载即净。
Installed cleanly when we ran it
tinqiao-oss
The Engramory curated-memory discipline as an installable plugin ([npm: dsh-engramory](https://www.npmjs.com/package/dsh-engramory)): a deterministic 200-line / 25 KB cap on the `MEMORY.md` index via `ctx.tools.guard()` — growth denied, a shrinking rewrite always passes — plus the protocol registered as a runtime skill. The store is plain markdown, one file per fact, shared with Claude Code, Codex, Kiro, and OpenClaw.
Installed cleanly when we ran it
ZSeven-W
Noema long-term memory plugin for DSH: durable, inspectable agent memory with recall tools and a settings page.
Installed cleanly when we ran it
