Skip to main content
This guide walks through creating a managed OpenClaw agent on OpenComputer, connecting it to Telegram, and chatting with it. By the end you’ll have a personal AI agent running Claude via OpenRouter, reachable from your phone. OpenClaw is a multi-channel AI gateway with a rich plugin SDK, native MCP support, and hot-reload configuration. It supports Telegram, Discord, Slack, Signal, WhatsApp, and more out of the box.

Prerequisites

Install the CLI

Install the oc binary from the CLI installation guide.

Step 1: Create the agent

This creates the agent, provisions an OpenClaw sandbox with Node 22 and the OpenRouter LLM provider pre-configured, and starts the OpenClaw gateway. Check that the agent is running:
Wait for the status to show running before proceeding. Provisioning typically takes 20-30 seconds.

Step 2: Create a Telegram bot

Open this link to BotFather in Telegram (or search for @BotFather in the app). Then:
  1. Tap Start if it’s your first time
  2. Send /newbot
  3. Enter a display name when prompted (e.g. “My OpenClaw Agent”)
  4. Enter a username — must end in bot (e.g. my_claw_test_bot)
  5. BotFather replies with a token like 110201543:AAHdqTcvCH1vGWJxfSeofSAs0K5PALDsaw — copy it

Step 3: Connect Telegram

The CLI will prompt you to paste the bot token. Alternatively, use the API directly:
This registers a webhook with Telegram and configures OpenClaw’s built-in Telegram channel. OpenClaw hot-reloads the config automatically — no gateway restart needed. Verify it worked:

Step 4: Chat with your agent

Open Telegram, find your bot by its username, and send a message. The agent responds using Claude (via OpenRouter).

Shell access

You can shell into the agent’s sandbox at any time:
This opens an interactive terminal inside the agent’s environment. Some useful commands:
See the OpenClaw docs for the full feature set: plugins, canvas, voice, browser tools, and more.

Cleanup

Disconnect the channel and delete the agent:
Disconnecting removes the Telegram channel config. Deleting the agent tears down the sandbox.

OpenClaw vs Hermes

Both cores support the same user journey (create / connect / install). The differences are under the hood: Choose Hermes for its self-improving skill loop and built-in memory. Choose OpenClaw for its breadth of channels and native plugin ecosystem.

What’s next

  • Install gbrain for persistent memory: oc agent install my-claw gbrain
  • Create a Hermes Agent — try the other managed core
  • Custom cores — bring your own agent runtime with oc shell and the raw sandbox API