snowfoxHQ/dsh-hmr--plugins-dsh-hmr ↗★ 0

dsh-hmr

dsh-Hestia Memory Runtime (dsh-hmr): cognitive memory for DeepSeek Harness — reasoning chains, reflection and self-evolution. Not hot module replacement. 适合需要为智能体引入推理链、反思与认知记忆能力的开发者。

Package
dsh-hmr
Compatibility
Unverified
Harness peer range
^0.2.1-rc.2
Cordis peer range
^4.0.5
Version
0.1.0
License
MIT
Last updated
Oct 20, 2026

Install

$npx -p @deepseek-ai/dsh dsh plugin --profile web add github:snowfoxHQ/dsh-hmr#750c474efb70039602f654efa653719a2231369e&path:plugins/dsh-hmr

Configuration

Everything is optional; the defaults work on a clean machine.

OptionDefaultMeaning
process'managed'managed spawns and supervises the Python service; external connects to one you run yourself.
baseUrlhttp://127.0.0.1:8077Service base URL. Host/port are reused when spawning.
token–Shared secret, sent as x-hmr-token / HMR_TOKEN.
python'python'Interpreter used to spawn the service.
serviceDirbundled copyDirectory containing server.py.
storagePath~/.dsh-hmr/dataWhere HMR keeps its data. Invalid paths fall back to the default with a warning.
startupTimeoutMs180000How long to wait for /health on first startup.
requestTimeoutMs120000Per-request timeout.
offlinefalseSets HF_HUB_OFFLINE=1 / TRANSFORMERS_OFFLINE=1 for the service. Turn this on when the embedding model is already cached — it cuts startup from ~75s to a few seconds.

Example cordis.patch.yml override:

- id: hmr-memory
  config:
    process: external
    baseUrl: http://127.0.0.1:8077
    token: !env HMR_TOKEN
    offline: true

Embeddings

The embedding backend is not configured here. HMR reads it from its own environment (HMR_EMBEDDING_*, .env next to the service). The plugin only forwards HMR_HOST, HMR_PORT, HMR_STORAGE_PATH, HMR_TOKEN and the offline switches.