Introduction

In intelligent agent workflows like DSH (DeepSeek Harness) where “everything is a plugin,” conversation memory often faces two pain points: first, contamination of the model-visible content with thinking chains, tool noise, and system injections; second, the lack of stable session and peer boundaries across multiple DSH conversations and multiple assistant perspectives.

Below is an introduction to dsh-honcho-memory. It addresses these two issues by providing Honcho v3 memory adaptation and shared knowledge base capabilities.

What is it

dsh-honcho-memory is a DSH plugin maintained by kingcharleslzy-ai and licensed under MIT. Its positioning is as a DeepSeek Harness (DSH) Honcho v3 memory adapter and shared knowledge base plugin.

It is necessary to clarify the boundaries first: this npm package is not the Honcho backend and will not install Honcho for you. Before use, you must have a Honcho v3 service accessible by DSH. Memory will not work if the Honcho API/Database/Backend deriver is missing.

Core Features

The following lists verified capabilities:

  • Automatically saves real user messages and model-visible responses, filtering out thinking chains, tool noise, and system injections.
  • By default, maps each DSH conversation to an independent Honcho session.
  • Recalls session summary, user/model, peer card, local conclusion, and shared knowledge in every turn.
  • Provides seven DSH tools:
memory_store
memory_search
memory_context
memory_reason
memory_profile
memory_dream
memory_status
  • Preserves directional perspectives (e.g., deepseek -> user) and does not masquerade different assistants as the same peer.
  • Uses shared-knowledge -> user/shared-knowledge as an optional canonical shared layer.
  • Deduplicates similar conclusions only upon reading; organization tools default to a dry-run and will not silently delete backend data.
  • Uses the Honcho official v3 HTTP API, targeting Node.js 20 or higher.
  • The installation package includes dsh-honcho-memory-core with zero runtime dependencies, so there is no need to install the core separately.

Prepare Honcho First

Below are two prerequisite options; choose one.

Option A: Use Honcho Official Hosting Service

When using the official hosted service, point the plugin configuration to the official API:

baseUrl: https://api.honcho.dev
apiKey: YOUR_HONCHO_API_KEY

Option B: Self-host Official Honcho

Deploy the official Honcho via Docker according to the Plastic Labs official repository. After deployment, perform a health check; the result should return:

{"status":"ok"}

If Honcho is running on another machine, ensure the DSH process can access that address, and protect the network link using HTTPS, VPN, or SSH tunneling; do not expose the unauthenticated Honcho port directly to the public internet.

Installation & Enablement

The installation command example is as follows:

dsh plugin --profile web add dsh-honcho-memory

Here, --profile web is an example provided in the documentation; whether other DSH profiles are applicable needs to be determined based on your profile configuration.

After installation, override the configuration in the cordis.patch.yml of the corresponding DSH profile. At a minimum, you should confirm that baseUrl, workspace, userPeer, and aiPeer point to the Honcho service and peer agreement you are actually using.

Check After Enablement

After restarting DSH, call:

memory_status({"check":"health"})

Then confirm the following items:

  1. The Honcho API is available.
  2. workspace, userPeer, and aiPeer match your configuration.
  3. The queue moves from pending/in-progress to completed.
  4. New conversations can be written using memory_store and retrieved using memory_search.

Tool Usage Notes

Among the seven tools, memory_status is used for status checks; memory_store and memory_search are used for write and read verification.

memory_dream may consume backend LLM resources and must explicitly state:

confirm=true

Data & Privacy

  • The plugin sends DSH conversations (within the enabled scope) to the baseUrl you configured.
  • The plugin itself does not contain maintainer account details, server addresses, or default user identities, nor does it provide telemetry services.
  • The Honcho backend may send content to its configured embedding, summary, deriver, or dialectic model providers; the privacy boundary depends on your own Honcho deployment and model configuration.
  • The plugin runs within the current DSH process; therefore, you should check the source code and the MIT license before installation and confirm that you are willing to send conversations to the selected Honcho service.

Upgrade & Compatibility

  • The target API is Honcho v3.
  • Node.js requirement is 20 or higher.
  • Starting from 0.5.2, defaults do not include maintainer identity or machine configuration.
  • Before upgrading from version 0.5.1 or earlier, you must confirm the actual location of old data and add baseUrl, workspace, and userPeer to your own cordis.patch.yml.
  • Version 0.5.2 will not migrate, rename, or delete existing Honcho data; only by continuing to use the same backend, workspace, and peer ID will original memories be recalled.
  • This plugin is not an official DSH integration released by Plastic Labs; installation and configuration for other clients like Codex or Hermes are not handled by this plugin.

Development & Verification

Development verification can be executed using the following commands:

npm install
npm test
npm run audit:public
npm run smoke
npm run smoke:dsh
npm run tidy:messages

Note that tidy:messages is consistent with the organization tool, defaults to dry-run, and will not silently delete backend data.

Use Cases & Notes

Suitable for developers who are already using Honcho v3 in DSH or are planning to self-host Honcho. It is suitable for users who need to connect DSH conversations to an independent memory backend while preserving peer perspectives and shared knowledge layers.

Notes:

  • The plugin runs with the permissions of the current DSH process and will send conversations (within the enabled scope) to the configured baseUrl.
  • You should check the source code and license before installation.
  • Memory will not work if only the plugin is installed without a complete Honcho v3 service.

Conclusion

The value of dsh-honcho-memory lies in connecting DSH’s messages, sessions, peer perspectives, and shared knowledge to Honcho v3, while leaving backend deployment, privacy boundaries, and upgrade data locations for you to confirm.

GitHub:

https://github.com/kingcharleslzy-ai/dsh-honcho-memory

The community directory page address was not provided in the verified materials; the DSH community directory is an independent site with no official affiliation with DeepSeek / Hypi, and you can search for dsh-honcho-memory in the directory.