OpenClaw

OpenClaw is a personal AI assistant you run on your own devices, one agent, reachable across WhatsApp, Telegram, Slack, Discord and more. The catch with a multi-channel agent is that memory usually fragments per channel. Korely fixes that: install one plugin, select it as your memory backend, and the agent carries the same memory, your typed facts, your graph, into every conversation.

korely-openclaw plugs into OpenClaw's exclusive memory slot. Once selected, your agent gets the memory_* tools backed by Korely, plus automatic recall before each turn.

Requirements

You also need a Korely API key. Get one free, the hobby tier is free, no credit card.

Install

openclaw plugins install korely-openclaw

Select it as your memory backend

OpenClaw uses one exclusive memory slot. Point it at korely and add your key (prefer ${KORELY_API_KEY} or a SecretRef over a plaintext value), then restart the gateway:

{
  plugins: {
    slots: { memory: "korely" },
    entries: {
      korely: {
        enabled: true,
        config: {
          apiKey: "${KORELY_API_KEY}",
          userId: "default",
          autoRecall: true,    // inject memory before each turn (default true)
          autoCapture: false,  // store the reply after each turn (default false)
          topK: 6,
        },
      },
    },
  },
}

openclaw gateway restart

What your agent gets

With Korely selected as the memory backend, the agent can:

• Use the memory_* tools, memory_search, memory_add, memory_get, memory_list, memory_update, memory_delete. These call Korely directly; typed facts are extracted and contradictions resolved server-side.
• Auto-recall, before each turn, Korely's get_context assembles the user's active typed facts plus the most relevant memories and injects them as a <korely-memory> block. Pure retrieval: no model runs on the read path.
• Auto-capture (opt-in), when enabled, the agent's reply is stored after each successful turn, embedded, and its entities extracted into the graph.

Example session: WhatsApp

Here is auto-recall in practice. You message your agent on WhatsApp; before the model runs, Korely injects the relevant context, and the agent answers with your own history.

You › When is my lease renewal due, and what did the landlord
say about the rent?

# auto-recall: get_context runs before the model and injects a
# <korely-memory> block with the facts that are true now
← injected context:

   "Lease renewal call with Anna (2026-05-28): Renewal deadline is
   July 1. Anna confirmed a 3% increase, 1,200 to 1,236 EUR per
   month starting August. She wants the signed copy by post."

Agent › Your lease renewal is due July 1. On the May 28 call Anna
confirmed a 3% increase, 1,236 EUR per month starting in August,
and she asked for the signed copy by post. Want me to create a
task with a reminder for the last week of June?

Read-only vs read-write

A multi-channel agent anyone can message is a bigger surface than a coding assistant on your laptop. That's why autoCapture is off by default: the agent reads memory (auto-recall + memory_search) but only writes when it explicitly calls memory_add. Once you trust the loop, turn on auto-capture to have replies stored automatically:

config: {
  apiKey: "${KORELY_API_KEY}",
  autoRecall: true,
  autoCapture: true,   // now replies are stored after each successful turn
}

Configuration

Troubleshooting

Why Korely

For the full method surface, see the SDK reference. For how facts, supersession and time work, see temporal facts.