Cortex Native Plugin For OpenClaw

This directory contains the real OpenClaw-native Cortex plugin package scaffold:

• package.json
• openclaw.plugin.json
• config.schema.json
• src/index.js
• src/service.js
• src/hooks.js
• src/identity.js

What it does:

• calls a local cortexd REST API over loopback
• injects live routed Cortex context in before_prompt_build
• seeds per-user and per-thread memory in agent_end
• resolves the same person across Telegram, WhatsApp, Discord, Slack, web chat, and similar channels
• keeps context self-hosted and user-owned

Start the local Cortex API first:

cortexd --store-dir ~/.openclaw/cortex --host 127.0.0.1 --port 8766

Then install from a local packed tarball:

cd examples/openclaw-plugin
TARBALL="$(npm pack --silent)"
openclaw plugins install "./$TARBALL" --force
openclaw plugins enable cortexai-openclaw
openclaw gateway restart

Once published, the install UX becomes:

openclaw plugins install cortexai-openclaw
openclaw plugins enable cortexai-openclaw
openclaw gateway restart

Version 1.6.1 does not launch child processes from inside OpenClaw. Running cortexd separately keeps the plugin install path scanner-friendly while preserving the same local-first memory behavior.

Recommended config:

{
  plugins: {
    entries: {
      "cortexai-openclaw": {
        enabled: true,
        hooks: {
          allowPromptInjection: true,
          allowConversationAccess: true
        },
        config: {
          apiBaseUrl: "http://127.0.0.1:8766",
          storeDir: "~/.openclaw/cortex",
          defaultTarget: "chatgpt",
          smartRouting: true,
          autoSeedThreads: true,
          projectDirStrategy: "agent-workspace",
          maxContextChars: 1500,
          failOpen: true
        }
      }
    }
  }
}

OpenClaw handles channel delivery and agent execution. Cortex handles portable context, cross-channel identity, and durable memory.