Skip to content

Self-evolving Context Database for AI Agents. Unify Agent Memory, Knowledge RAG and Skills.

4.3 KB
Raw Blame

OpenViking Memory

Long-term memory for DeepSeek Harness, backed by an OpenViking server. Works in DeepSeek Harness Desktop and in dsh web.

  • Recall: before each step that takes your input, relevant memories are added to the conversation.
  • Capture: your messages and the assistant's replies are sent to OpenViking and committed as memories.
  • Tools and skills: the model gets the OpenViking tools (mcp__openviking__search, read, remember, add_skill, …) and the openviking-memory, openviking-skills and ov-experience-memory skills.
  • URI protection: a file tool aimed at a viking:// path is stopped and pointed at the right OpenViking tool. A bash or pwsh command that contains one still runs, with a note to the model.

1. Connect to your server

Write ~/.openviking/ovcli.conf with url and api_key (plus account and user on a trusted-mode server). A local server at http://127.0.0.1:1933 without authentication needs nothing. See Deployment Guide → CLI.

Prefer ovcli.conf to environment variables in Desktop. On macOS, Desktop reads your login shell's environment once when it starts; on Windows, it uses your user environment. After changing either, fully quit and reopen Desktop.

2. Install

With the installer (macOS):

curl -fsSL https://openviking.ai/install | bash

It needs Node.js 18 or later. It finds Desktop and installs into it as the dsh-desktop harness. Open Desktop once beforehand so its profile exists, and quit it with ⌘Q while the installer runs: closing the window only hides it.

From the Plugins page (manual; macOS, Windows and dsh web): Plugins → Add plugin → enter @openviking/dsh-memory-plugin → Install, then turn it on if its switch is off. This only adds the plugin: it does not ask for your OpenViking credentials, so do step 1 first.

With Desktop's own dsh command (macOS): open Desktop once, quit it with ⌘Q, then run:

"/Applications/DeepSeek Harness.app/Contents/Resources/runtime/cli/bin/dsh" plugin --profile desktop add @openviking/dsh-memory-plugin

A dsh installed from npm cannot manage the desktop profile. After a command-line install, reopen Desktop.

3. Check that it works

Open a new conversation. You should see an OpenViking context block near the top, and the model should list mcp__openviking__* tools. Ask about something from an earlier conversation to confirm recall.

To check the install from a terminal (with Desktop quit):

"/Applications/DeepSeek Harness.app/Contents/Resources/runtime/cli/bin/dsh" plugin --profile desktop ls

Good to know

  • No settings page. Configure the plugin through ovcli.conf or OPENVIKING_* variables. Behavior options can also go in ~/.dsh/profiles/desktop/cordis.patch.yml under openviking-memory-runtime. The recovery dialog's "Disable third-party plugins, back up profile patch, and restart" resets the profile's plugins and moves that file aside, so keep lasting settings in ovcli.conf, and enable the plugin again afterwards.
  • Tool identity. Automatic recall and capture follow each conversation's project. The OpenViking tools run once per profile, from the profile directory, so with OPENVIKING_RECALL_PEER_SCOPE=actor they have no project peer. To give them one, set actor_peer_id in ovcli.conf or OPENVIKING_PEER_ID. That peer also replaces each conversation's project peer for recall and capture, so every conversation shares it.
  • Skills. Bundled skills reach agents whose preset includes the skill tool. The minimal preset does not, so agents using it do not see them.
  • Logs. Desktop shows no plugin output. To debug the OpenViking tools, set OPENVIKING_DEBUG=1 and OPENVIKING_DEBUG_LOG=<file>, then reopen Desktop.

Uninstall

Plugins → OpenViking Memory → Uninstall. Or, with Desktop quit:

"/Applications/DeepSeek Harness.app/Contents/Resources/runtime/cli/bin/dsh" plugin --profile desktop rm @openviking/dsh-memory-plugin

Full documentation: DeepSeek Harness integration and the bundle README.