Connects DeepSeek Harness to ReMe's local-first, self-evolving personal knowledge base: automatically captures completed main-agent conversations as user-owned Markdown memory, searches conversations and source material through reme_search with BM25, optional embeddings, and wikilink expansion, and schedules daily memory consolidation.
Install
# from npm (prebuilt)
dsh plugin --profile web add @agentscope-ai/reme
# from GitHub (first run asks for allowBuilds approval — follow the hint, retry)
dsh plugin --profile web add github:agentscope-ai/ReMe#path:/typescript
Any plugin you install runs third-party code with your own permissions — it can read your files, use your credentials, and reach the network, and tool approvals don’t sandbox it. GitHub-sourced plugins also run build scripts at install time — pnpm blocks those until you allow them, so an install can stop with ERR_PNPM_GIT_DEP_PREPARE_NOT_ALLOWED or ERR_PNPM_IGNORED_BUILDS; dsh prints the exact key to add under allowBuilds in your profile’s pnpm-workspace.yaml, and the install works on the next run. Allowing a build is a trust decision: only install sources you trust, and pin a commit (github:owner/repo#sha).
README
@agentscope-ai/reme provides one shared ReMe HTTP client and host adapters for DeepSeek Harness and OpenClaw. Each
adapter uses its host's native lifecycle and tool interfaces; importing the root package does not load either host.
The package expects a running ReMe HTTP service with the search, auto_memory, and auto_dream jobs required by the
selected adapter:
reme start workspace_dir=/absolute/path/to/workspace
The default endpoint is http://127.0.0.1:2333. All entries support REME_URL, or REME_HOST plus REME_PORT.
ReMe's HTTP service does not use API-key authentication.
Requirements
- A running ReMe HTTP service with the jobs required by the selected adapter.
- Node.js
22.22.3+,24.15.0+, or25.9.0+on the corresponding supported major-version line. - DeepSeek Harness or OpenClaw only when using that host-specific entry; the root client has no host runtime dependency.
DeepSeek Harness
Install the package as a DSH profile bundle:
dsh plugin --profile web add @agentscope-ai/reme
The bundle loads @agentscope-ai/reme/dsh in an isolated remeMemory realm. It injects durable memory guidance,
registers reme_search, submits completed main-agent turns to auto_memory, and runs the optional daily auto_dream
schedule. Recalled plugin context and tool results are excluded from automatic memory capture.
Configure the bundle by replacing its row in the profile's cordis.patch.yml:
- id: reme-memory
config:
- id: reme-memory-runtime
name: "@agentscope-ai/reme/dsh"
config:
endpoint: http://127.0.0.1:2333
language: zh
timezone: Asia/Shanghai
autoMemoryInterval: 5
autoDreamEnabled: true
dreamCron: "0 23 * * *"
On the DSH Web profile, the same fields are available under Settings → Plugins → Plugin configuration → ReMe
Memory. Changes are stored in DSH's user settings document and apply to subsequent requests and captures. Changing
the daily dream controls reschedules the next run; changing the guidance language affects newly started sessions.
The test-only dreamIntervalMs value remains outside the user-settings section.
| Option | Default | Meaning |
|---|---|---|
endpoint |
http://127.0.0.1:2333 |
ReMe HTTP service URL |
language |
en |
Memory guidance language: en or zh |
autoMemoryEnabled |
true |
Capture completed main-agent turns |
autoMemoryInterval |
5 |
Submit after this many completed turns |
autoDreamEnabled |
true |
Enable daily dream maintenance |
dreamCron |
0 23 * * * |
Daily schedule in the workspace timezone |
dreamHint |
empty | Optional guidance sent to auto_dream |
rootAgentsOnly |
true |
Exclude subagents from guidance and capture |
searchLimit |
5 |
Maximum results returned by reme_search |
requestTimeoutMs |
10000 |
Search request timeout |
backgroundTimeoutMs |
3600000 |
Automatic-memory and dream timeout |
shutdownTimeoutMs |
5000 |
Best-effort shutdown drain budget |
timezone |
Asia/Shanghai |
IANA timezone used for batches and scheduling |
The ReMe card reads the service's health_check and status jobs on demand. It shows the ReMe version, component
health, chunk/index counts, process RSS, and estimated component memory; it can also display the redacted app_config
response and trigger one auto_dream run. Diagnostics are refreshed when the card first opens or when the user asks,
not polled continuously. The page calls the configured ReMe HTTP endpoint from the local browser, so that service must
remain browser-reachable and allow the DSH origin.
OpenClaw
OpenClaw 2026.7.1 or later can install the same package. The current SDK and
OpenClaw Gateway require Node.js 22.22.3+, 24.15.0+, or 25.9.0+ on their
respective major-version lines:
openclaw plugins install @agentscope-ai/reme
Select reme for plugins.slots.memory when another memory plugin is active. The adapter uses OpenClaw's current
before_prompt_build hook, registers the reme_search action, injects durable memory guidance, and recalls relevant
memory before conversational root-agent runs. Completed user/assistant pairs are grouped into per-session,
date-consistent batches for auto_memory; failed batches are retained for retry and pending work is flushed within a
bounded Gateway shutdown budget. One plugin-owned daily schedule runs auto_dream. Recall is wrapped in
<reme-context> and marked as untrusted historical data. Cron, heartbeat, memory, overflow, and subagent runs do not
recall or capture conversational memory by default.
OpenClaw plugin configuration accepts:
| Option | Default | Meaning |
|---|---|---|
endpoint |
http://127.0.0.1:2333 |
ReMe HTTP service URL |
language |
en |
Memory guidance language: en or zh |
autoRecall |
true |
Recall before conversational root-agent runs |
searchLimit |
5 |
Maximum search results |
recallMinScore |
0 |
Minimum search score |
autoMemoryEnabled |
true |
Capture completed conversational turns |
autoMemoryInterval |
5 |
Submit after this many completed turns |
autoDreamEnabled |
true |
Enable daily memory consolidation |
dreamCron |
0 23 * * * |
Daily schedule in the workspace timezone |
dreamHint |
empty | Optional guidance sent to auto_dream |
rootAgentsOnly |
true |
Exclude subagents from guidance and capture |
timezone |
Asia/Shanghai |
IANA timezone used for batches and scheduling |
requestTimeoutMs |
10000 |
Recall and explicit search timeout |
backgroundTimeoutMs |
3600000 |
Automatic-memory and dream timeout |
shutdownTimeoutMs |
5000 |
Best-effort shutdown drain budget |
OpenClaw's conversation-access and prompt-injection permissions remain host settings; enable them for ReMe when your OpenClaw policy requires explicit grants. The adapter does not modify Gateway configuration.
Library entry
Consumers that only need the transport can import the root package:
import { ReMeClient, formatReMeContext } from "@agentscope-ai/reme";
Host code is available only through @agentscope-ai/reme/dsh and @agentscope-ai/reme/openclaw.
Development
cd typescript
npm ci
npm run format:check
npm run lint
npm run typecheck
npm test
npm run test:package
npm pack runs the TypeScript build and includes only dist, the DSH patch, the OpenClaw manifest, and the English and
Chinese READMEs.
Stable releases use npm's latest distribution tag; prereleases use next.
Links
More in this category
volcengine/OpenViking#examples/dsh-memory-plugin★ 34580
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.
vectorize-io/hindsight#coding-agents★ 21937
Hindsight, agent memory that learns: long-term project memory with auto recall and retain, knowledge pages, deep reflection, and per-repo memory banks.
zilliztech/memsearch#MemSearch★ 2533
Shared Markdown memory for DSH and other coding agents, with automatic capture, pre-step context injection, searchable recall, and memory-to-skill self-evolution through a review panel.
vshulcz/deja-vu#extensions/dsh★ 742
Reads the session files nineteen other coding agents on this machine already wrote — Claude Code, Codex, Cursor, opencode, Antigravity, Kimi, Cline, Zed and more — including sessions from before it was installed: six tools (deja_recall, deja_session, deja_blame, deja_fix, deja_how, deja_remember), a /deja command, and optional automatic recall added to the runtime context. Local BM25 index, no LLM, no embeddings, no network (dsh plugin --profile web add dsh-deja).
adoresever/graph-memory★ 586
Traceable, searchable cross-session memory for DeepSeek Harness — conversation knowledge as typed graph nodes (TASK/SKILL/EVENT) and typed edges.
Ikalus1988/MisakaNet★ 432
Failure-recovery memory: search and record failure-recovery lessons from real engineering sessions, with BM25 + semantic RAG retrieval and a lessons knowledge base.
Community comments
Comments are public GitHub Discussions. Loading them connects to GitHub and Giscus; a GitHub account is required to post.