memorylake-harness

Client-side harnesses that connect coding agents to MemoryLake — cross-device long-term memory. (Two deployments, separate accounts: memorylake.ai international, memorylake.cn China; the CLI defaults to the international endpoint.)

All harnesses share one identity and data tree (~/.memorylake/harness/): configure once, use from every client.

Every harness drives the memorylake CLI under the hood and fetches it on demand during setup, so none of the installs below needs a separate download. Put the CLI on your PATH to reach the same memory from a terminal or a script — see From the terminal.

Claude Code

/plugin marketplace add memorylake-ai/memorylake-harness
/plugin install memorylake@memorylake
/memorylake:init

See claude-plugin/README.md for configuration, privacy notes, and design rationale.

Codex

codex plugin marketplace add memorylake-ai/memorylake-harness
codex plugin add memorylake@memorylake
set up memorylake

The last line is asked inside a Codex session, after trusting the hooks via /hooks. See codex-plugin/README.md.

DeepSeek Harness (dsh)

dsh plugin --profile web add @memorylake/dsh-plugin
dsh web
/memorylake-init

The last line is typed in a session; machines already set up for Claude Code or Codex can skip it. See dsh-plugin/README.md.

opencode

opencode plugin @memorylake/opencode-plugin -g
set up memorylake

The last line is typed in a session; machines already set up for another harness can skip it. See opencode-plugin/README.md.

QwenPaw

qwenpaw plugin install https://github.com/memorylake-ai/memorylake-harness/releases/download/qwenpaw-plugin-v0.2.2/memory-memorylake-0.2.2.zip

Then, per Agent, in the Console: Agent settings → Memory backend → Memory Lake. On a machine already set up for another harness the form can stay empty. See qwenpaw-plugin/README.md.

QwenWork

In a QwenWork conversation, send:

Install the MemoryLake plugin by following qwenwork-plugin/INSTALL.md in https://github.com/memorylake-ai/memorylake-harness

The agent installs the plugin and runs the setup; machines already set up for another harness skip the setup. Agents installing into QwenWork: follow qwenwork-plugin/INSTALL.md -- the plugin is that directory, not the repository root. See qwenwork-plugin/README.md.

WorkBuddy

In WorkBuddy: Skills → Plugins → + (add marketplace) → memorylake-ai/memorylake-harness, install memorylake, then in a session:

/memorylake:init

Machines already set up for another harness can skip the last step. See workbuddy-plugin/README.md.

From the terminal

The harnesses log in through the CLI and keep the credentials where the CLI keeps them (~/.memorylake/credentials.toml), so a CLI you install yourself on a machine already set up for any harness is logged in from the first command:

curl -fsSL https://raw.githubusercontent.com/memorylake-ai/memorylake-cli/main/scripts/install.sh | sh
memorylake auth status
memorylake lib upload ./report.pdf
memorylake search "what did we decide about the schema"

Windows: irm https://raw.githubusercontent.com/memorylake-ai/memorylake-cli/main/scripts/install.ps1 | iex. The full command reference — files, facts, databases, conversations, agents, skills, teams — is in the memorylake-cli README.

Beyond coding sessions

The memory the harnesses write to is the same one a team's calls, meetings, documents and agents can land in. memorylake-usecases has runnable demos of that — sales calls, meetings, a codebase, contract review, bookkeeping, agent bug replay, Reflexion-style agent memory and more — each a directory with sample data, a CLI runner that prints every command, a local web UI and a recorded run. A machine set up for any harness can run them as is: the demos use the CLI the harness installed and, with MEMORYLAKE_API_KEY unset, its login.