Install Context Mode Engine in your agent

First, check that you have Node 24.15.0 or later: run node -v. Each coding agent installs plugins its own way, so a new tool can feel strange. Context Mode Engine follows the way your agent already uses: run the agent's own install command, restart the agent, then run one check. Ten agents, one page, and every command comes from the plugin's own install guide.

Free · Local · Private · Yours to keep

Context Mode Engine is source-available under the Elastic License 2.0. Use it, copy it, change it and share it at zero cost, free for all time. It runs on your machine, sends zero usage data, and keeps your data in ~/.context-mode.

Every agent installs in its own way

Claude Code uses a plugin marketplace. Gemini CLI uses extensions. Pi installs from npm. When each tool asks for a different config file, setup takes longer than the work.

Context Mode Engine uses the command your agent already knows. You type what you type for any other plugin in that agent. The Engine then brings its tools, hooks and skills with it.

Before you start

Check your Node version first. The Engine needs Node 24.15.0 or later, because it uses the built-in node:sqlite:

• On Windows: Git for Windows. The shell tools run in Git Bash, and context-mode doctor checks it.
• One of the ten agents on this page. To see which ones work on your machine, run context-mode setup --list.

The context-mode command

Codex trust in step 1 and the check in step 3 use the context-mode command in your terminal. Install it once, before step 1:

Step 1: pick your agent and run its command

Pick your agent below. Copy its install command and run it in your terminal. The agents fall in three groups: agents with a plugin system, agents that load the Engine in process (Pi and OpenCode), and agents that use npm and context-mode setup.

Why the commands end in different tags: each agent has its own way to name a branch on GitHub. Claude Code writes mksglu/context-mode#release, Codex writes mksglu/context-mode@release, and Gemini CLI and Qwen Code add --ref release. All four point to the same release branch, in the form each agent's own command expects. Agents that install from npm (Pi, OpenCode and the npm group) name the next tag on npm with @next.

The same commands in one table:

For the npm group, the <id> is copilot-cli, kiro, vscode or zcode. Source: the plugin's install guide.

Codex: trust the hooks

Codex runs a hook after you trust it. After codex plugin add, Codex asks you to trust the hooks; setup marks them trusted. Run it once:

Step 2: restart the agent

Close your agent and open it again. The agent loads the Engine's tools, hooks and skills when it starts. Restart it again after each update.

Step 3: check the install

Inside the agent, run /ctx-doctor. In Pi, or in any terminal, run:

Doctor checks Node, the tools, the hooks and the skills for each agent. Each check shows PASS when it is ready. If a check asks for a fix, doctor prints the command to run. To repair a setup, run context-mode setup.

What you get

Your agent now sends large tool output to a local sandbox and index, and reads a short reply. The hooks stay light. In the 2026-10-31 run, a hook added 7.3 ms to 14.6 ms per tool call, by agent. That is the median of 5 hook runs, minus the median of 5 bare node starts (raw results).

Next, run your first session and see the bytes the Engine keeps out of your context.