DSH Plugins Marketplace

DSH Plugins

Plugins

/

dsh-onto-space-panel

h

dsh-onto-space-panel

Manifest valid

Onto space panel for DeepSeek Harness: declarations, agents, memory and per-chat context.

UI (client)hasBundlePatch

Onto Space Panel for DeepSeek Harness

A compact Onto space picker and a side panel for declarations, resident agents, memory artifacts, templates and objects. Select what belongs in the current chat; the active context remains visible above the composer.

Features

  • Connect to Onto on first use: enter your MCP endpoint and API key, verify access, then select a space. Reuse an existing Harness connection automatically.
  • Search and select available Onto spaces.
  • Open the space panel with Обзор.
  • Read an accepted/current space declaration and attach it to the chat.
  • Inspect resident agents and their charters. Selecting a valid active resident loads its accepted charter and constitution into the current chat.
  • Search and page through accepted memory artifacts, read their contents and explicitly attach or disconnect them.
  • Browse templates and objects, then include selected objects in the conversation.
  • Keep selections and context separate for each chat and each selected realm.
  • Change context through the select_onto_space and configure_onto_chat_context model tools when the user requests it.

The UI is currently in Russian. Names and descriptions are supplied by Onto. An agent selection gives the current Harness conversation a resident role; it does not launch a separate agent process or external agent broker.

Requirements

  • DeepSeek Harness with its Web UI or Desktop application.
  • Your Onto MCP endpoint URL and personal API key, or an existing authenticated Harness MCP connection named onto.
  • Harness's writable credential provider for the first-use setup form. Standard local profiles use @deepseek-ai/dsh-credentials-local.
  • The connection must expose list_available_realms, about_realm, list_realm_agents, get_realm_agent, search_memory_artifacts, get_memory_artifact, get_memory_artifact_by_path, search_templates, search_objects and get_entity for their respective panel features.

Host integration was developed against DeepSeek Harness 0.2.0-rc.2. Other releases are not yet verified. Harness is in developer preview and its extension interfaces can change.

Install a release tarball

Download the package .tgz from the project's release and install it into a Web-capable profile. For example:

dsh plugin --profile web add ./dsh-onto-space-panel-1.2.1.tgz

Release downloads: https://github.com/hope4b/dsh-onto-space-panel/releases

You can also install the tagged GitHub source, which already contains the built browser entry point and has no install-time build script:

dsh plugin --profile web add github:hope4b/dsh-onto-space-panel#v1.2.1

In Desktop, use Plugins to install the downloaded bundle where supported by your version. The tarball already contains the browser bundle; no install-time build script or extra build permission is needed.

If this plugin was previously loaded manually by an absolute-path row in cordis.patch.yml, disable that old row before enabling the packaged bundle. Do not run two copies: they register the same tools and API routes.

Restart Harness after installing or replacing the bundle. Select an Onto space above the composer and press Обзор to open its panel. Use В контекст to attach a declaration or memory artifact. A role or material can be removed using the × next to its context label.

Connect Onto MCP

First-use setup (recommended)

  1. Install the plugin and restart Harness.
  2. Click Onto → Подключить above the composer.
  3. Enter your MCP endpoint URL and personal API key, then press Подключить. The key field expects only the value. A pasted X-Onto-Api-Key: value line is accepted too: the plugin removes the header name before sending or storing it.
  4. The plugin checks access with list_available_realms before saving. After success, select a space from the list. An empty list means the connection works but no accessible spaces were returned.

Obtain the endpoint and key from your Onto administrator. The form uses HTTPS and the X-Onto-Api-Key header; localhost HTTP is supported for development. It requires a direct MCP endpoint, without credentials or query parameters in the URL. Authentication redirects are rejected.

The connection belongs to the active Harness profile. Keys are stored through Harness's credential provider, not in plugin configuration, browser storage or chat messages. With the standard local provider, the managed credential file is $DSH_HOME/.credentials.yaml. This is local credential storage, not a claim of OS keychain encryption. The server address and credential reference are stored under $DSH_HOME/storages/onto-connections/, keyed by the profile location. The plugin restores its own MCP connection on startup without editing profile YAML.

Open Подключение in the space picker, or Подключение Onto in the side panel, to update a connection created by the plugin. A blank key reuses the saved key only when the endpoint stays the same. Changing the endpoint or account key requires choosing spaces and attaching context again; old chat context is not automatically applied to the new connection.

If Harness already provides a server named onto, the plugin reuses it and skips onboarding. Those external settings remain managed in Harness; the panel does not overwrite or migrate that connection. A configured but unavailable connection is shown as configured, with errors when loading spaces.

Manual Harness configuration (fallback)

Use this when your deployment manages MCP centrally or does not provide a writable credential provider. If the profile already has an Onto connection, do not add a second one.

  1. Obtain your Onto MCP endpoint URL and personal API key from your Onto administrator. Use the MCP endpoint, not the ordinary Onto website URL.
  2. Open the active profile's cordis.patch.yml: normally ~/.dsh/profiles/web/cordis.patch.yml for Web or %USERPROFILE%\.dsh\profiles\desktop\cordis.patch.yml for Windows Desktop. A custom DSH_HOME changes the .dsh location.
  3. Append the following patch, replacing the placeholder URL. If an Onto entry already exists, update it instead of creating a duplicate. The server name must be exactly onto.
- insert:
    - id: mcp-onto
      name: '@deepseek-ai/dsh-mcp-client'
      config:
        serverName: onto
        transport: streamable-http
        url: https://YOUR-ONTO-MCP-HOST/mcp
        headers:
          X-Onto-Api-Key: !!js process.env.ONTO_API_KEY

This example uses Onto's X-Onto-Api-Key header. Confirm the authentication scheme with your server administrator if your deployment differs. A copy is available in examples/onto-mcp.patch.yml.

  1. Set ONTO_API_KEY in the environment of the process that starts Harness. For Web on Windows, start it from the same PowerShell window:
$env:ONTO_API_KEY = 'YOUR_PERSONAL_API_KEY'
dsh --profile web

For Windows Desktop, fully quit Harness first, set the variable above, then launch the executable from that same PowerShell window. For the default installer location:

& "$env:LOCALAPPDATA\Programs\DeepSeek Harness\DeepSeek Harness.exe"

Launching via an existing desktop shortcut does not inherit a variable set only in this PowerShell window. The variable must be available on every launch. Enter your real key locally; do not commit it or share terminal history containing it.

  1. Open the space selector above the composer. Your accessible spaces should appear. You can also ask the model to call mcp__onto__list_available_realms.

If connection fails, check the endpoint and network access first. A 401 or 403 response usually indicates authentication or access rights. If the tools work in chat but the panel cannot find them, check that serverName is onto. Missing agent or memory tools affect those sections separately; an empty list can also mean your account has no access to those resources.

The transport and environment-variable syntax follow the official Harness MCP client configuration.

Privacy and context

The plugin makes read-only Onto calls through Harness's MCP client. It does not publish Onto data or perform artifact lifecycle mutations. Attached documents are sent to the model used by your Harness chat as context.

Selections and document snapshots are persisted locally under:

$DSH_HOME/storages/onto-space-selections/
$DSH_HOME/storages/onto-chat-contexts/

When DSH_HOME is unset, the usual ~/.dsh home is used. The context directory contains copies of documents you attached. It is not part of the distributed package. Context changes are protected against stale revisions and realm switches.

Missing declarations, unavailable agent governance and empty memory collections are displayed explicitly. An unavailable resident cannot be selected. Connected materials are snapshots from the time they were attached, not a live synchronization of their later revisions.

Package locally

The checked-in client.js is ready to load, so packaging requires no compilation:

npm run check
npm pack

From a source checkout, run the synthetic public tests with npm test.

The package uses Node built-ins on the host and Harness's shared React runtime in the browser. It has no runtime dependencies of its own and no install scripts.

Validation

The development checks covered real Onto response formats, declaration reading, resident validation and accepted governance loading, memory list/body reading, context projection, persistence, per-chat/per-realm isolation, stale revisions, and realm switches during attachment. Browser checks covered space selection, panel navigation, context attachment and separation between conversations.

Private example data and development fixtures are excluded from this release.

License

MIT. This is a community integration, not an official DeepSeek product.

Comments

Loading…

Similar plugins

dsh-butler-memory

by AndyYang12345

PostgreSQL-backed durable memory for DSH agents: MCP memory tools plus a Web panel for reviewing memories and candidates.

Terminal & ClientsMemory & ContextManifest valid

★ 0

Apache-2.0

TypeScript

Oct 5, 2026

dsh plugin --profile web add dsh-butler-memory

by jonah791

Agent-driven long-term memory for DeepSeek Harness: scoped memory (global + per-workspace), layered entries (fact/knowle

Manifest valid

★ 3

↓ 75/wk

MIT

TypeScript

Sep 7, 2026

dsh plugin --profile web add dsh-agent-memory

by elementor-i

agentmemory for DeepSeek Harness (dsh): full memory_* tools, capture hooks, and context injection over the local REST server

Tools & CapabilitiesSessions & MessagesManifest valid

★ 4

↓ 51/wk

MIT

TypeScript

Aug 17, 2026

dsh plugin --profile web add @elementor-i/dsh-agentmemory

by luyy9apples

Approval-gated workspace instructions and shared project memory for DeepSeek Harness

Manifest valid

★ 0

MIT

TypeScript

Sep 2, 2026

dsh plugin --profile web add dsh-workspace-memory

by guyuefangyuanl

Cross-session persistent memory for the DeepSeek Harness: a model-facing memory tool, an always-on index section, and a bundled maintenance skill.

Manifest valid

★ 3

↓ 545/wk

MIT

JavaScript

Aug 21, 2026

dsh plugin --profile web add dsh-memory

by perinchiang

DeepSeek Harness embedded read-only viewer for TencentDB Agent Memory's four-layer local memory

Sessions & MessagesManifest valid

★ 0

MIT

JavaScript

Aug 27, 2026

dsh plugin --profile web add dsh-memory-dashboard