OpenClaw: Build Your Personal AI Assistant

What You're Building

In this tutorial, you'll deploy a production-ready OpenClaw AI assistant. By the end, your assistant will be able to search and scrape the live web (Bright Data), remember conversations across sessions (Mem9 + TiDB Cloud Zero), reason with a Southeast-Asia-tuned LLM (Agnes AI), and chat with you on Telegram — all running on a single Zeabur Dedicated Server.

Prerequisites

• A Zeabur account (Dedicated Server plan — minimum 2 vCPU / 4 GB RAM)
• A Telegram account and a phone with the Telegram app installed
• A Bright Data account with an API key
• A Mem9 account for persistent memory
• A TiDB Cloud Zero database (no sign-up required)
• A ZenMux account for Agnes AI model access
• Basic comfort with environment variables and JSON files

Architecture Overview

┌────────────┐     ┌──────────────────────────────────────────────┐
│  Telegram   │◄───►│            OpenClaw (Zeabur)                 │
│  User Chat  │     │  ┌────────────┐  ┌────────────────────────┐ │
└────────────┘     │  │ LLM Router │  │  Skills                │ │
                    │  │ Agnes AI   │  │  ├─ Bright Data (Web)  │ │
                    │  │ GLM-4.7    │  │  ├─ Mem9 (Memory)      │ │
                    │  │ Grok 4     │  │  └─ TiDB (Database)    │ │
                    │  └────────────┘  └────────────────────────┘ │
                    └──────────────────────────────────────────────┘

OpenClaw — Your AI Assistant Framework

OpenClaw is an open-source AI assistant framework that runs on your own infrastructure. It acts as a gateway between you and various AI models, skills, and communication channels. Think of it as a personal chief of staff that's always online, always learning, and ready to handle whatever you throw at it.

Deploying OpenClaw on Zeabur

Open the one-click template and hit Deploy:

👉 https://zeabur.com/templates/VTZ4FX

You'll see a deployment form with three fields:

Choose Dedicated Server as the server type, select the OpenClaw destination, and click Confirm.

Deployment takes roughly 60–90 seconds. Once the status turns green, open your domain URL in a browser and log in with the OPENCLAW_GATEWAY_TOKEN shown in the Zeabur Instructions tab.

OpenClaw Commands Reference

Once your assistant is running, here are the essential commands you'll use:

Telegram Bot Integration

Your assistant needs a channel to talk through. Telegram provides a reliable, real-time messaging platform that works on any device.

Creating a Telegram Bot

Telegram bots are created through the official @BotFather service:

1. Open Telegram and message @BotFather
2. Send /newbot
3. Pick a display name (e.g. Agent Forge Bot)
4. Pick a username ending in bot (e.g. agentforge_hackathon_bot)
5. Copy the token BotFather gives you (looks like 123456789:ABCdef...)

Connecting the Bot to OpenClaw

Add the bot token to your Zeabur environment:

Go to your OpenClaw service in Zeabur → Variable tab → add:

TELEGRAM_BOT_TOKEN=123456789:ABCdefGHIjklMNOpqrsTUVwxyz

Restart the service. Then open the OpenClaw dashboard → Channels panel → click Connect Telegram Bot.

Your bot is now ready to receive and respond to messages.

Agnes AI and the Model Ecosystem

OpenClaw comes pre-configured with multiple AI models accessible through the Zeabur AI Hub:

Agnes AI via ZenMux

1. Sign up at zenmux.ai and create an API key
2. In Zeabur → Variable tab, add:

ZENMUX_API_KEY=your_zenmux_api_key_here

1. Open the Zeabur File Management tab and edit /home/node/.openclaw/openclaw.json. Replace the contents with this complete configuration:

{
  "models": {
    "providers": {
      "zenmux": {
        "baseUrl": "https://zenmux.ai/api/v1",
        "apiKey": "${ZENMUX_API_KEY}",
        "api": "openai-completions",
        "models": [
          {
            "id": "sapiens-ai/agnes-1.5-pro",
            "name": "Agnes 1.5 Pro (ZenMux)",
            "reasoning": false,
            "input": ["text"],
            "cost": {
              "input": 1.0,
              "output": 4.0,
              "cacheRead": 0.1,
              "cacheWrite": 0
            },
            "contextWindow": 128000,
            "maxTokens": 8192
          }
        ]
      },
      "nosana": {
        "baseUrl": "{BASE_URL}",
        "apiKey": "${NOSANA_API_KEY}",
        "api": "openai-completions",
        "models": [
          {
            "id": "{NOSANA_MODEL_ID}",
            "name": "GLM-4.7 Flash Q4_K_M (Nosana)",
            "reasoning": false,
            "input": ["text"],
            "cost": {
              "input": 0,
              "output": 0,
              "cacheRead": 0,
              "cacheWrite": 0
            },
            "contextWindow": 128000,
            "maxTokens": 8192
          }
        ]
      }
    }
  },
  "agents": {
    "defaults": {
      "model": {
        "primary": "zenmux/sapiens-ai/agnes-1.5-pro",
        "fallbacks": ["nosana/{NOSANA_MODEL_ID}"]
      },
      "models": {
        "zenmux/sapiens-ai/agnes-1.5-pro": {}
      }
    }
  },
  "commands": {
    "native": "auto",
    "nativeSkills": "auto",
    "restart": true,
    "ownerDisplay": "raw"
  },
  "channels": {
    "telegram": {
      "enabled": true,
      "dmPolicy": "pairing",
      "botToken": "${TELEGRAM_BOT_TOKEN}",
      "groups": {
        "*": {
          "requireMention": true
        }
      },
      "groupPolicy": "allowlist",
      "streaming": "partial"
    }
  },
  "gateway": {
    "controlUi": {
      "dangerouslyAllowHostHeaderOriginFallback": true,
      "allowInsecureAuth": true,
      "dangerouslyDisableDeviceAuth": true
    },
    "auth": {
      "mode": "token"
    },
    "trustedProxies": ["10.0.0.0/8", "172.16.0.0/12"]
  },
  "meta": {
    "lastTouchedVersion": "2026.3.13",
    "lastTouchedAt": "2026-03-22T06:27:58.962Z"
  }
}

Restart the service. Agnes 1.5 Pro now appears in your model list — switch to it with the /model command in Telegram.

Nosana — Decentralized GPU Compute

OpenClaw can connect to Nosana's inference endpoints using the same OpenAI-compatible gateway pattern. To use Nosana as a model provider:

1. Create an account at nosana.io and generate an API key
2. In Zeabur → Variable tab, add your Nosana credentials
3. Update your openclaw.json to include Nosana as a provider with the appropriate baseUrl and model IDs

This gives you another option for model inference alongside Zeabur AI Hub and ZenMux, letting you choose the infrastructure that best fits your needs.

Bright Data — Web Search & Scraping

Getting Bright Data Credentials

1. Log in to brightdata.com and go to Web Access → SERP API
2. Create a new zone (or use the default one) and copy the Zone name
3. Go to Account Settings → API Tokens and copy your API key

Installing the Bright Data Skill

Message your Telegram bot to install the Bright Data skill using natural language:

> "Install the Bright Data skill for web search."

OpenClaw will automatically fetch and install the skill from ClawHub.

Configuring Bright Data

In Zeabur → Variable tab, add:

BRIGHTDATA_API_KEY=your_bright_data_api_key
BRIGHTDATA_SERP_ZONE=your_zone_name

Restart the gateway by messaging your Telegram bot:

> /restart

Using Bright Data

Once configured, ask your bot to search the web:

> "Search for the latest AI hackathon news in Singapore and summarise the top 3 results."

The agent will use Bright Data's SERP API to search and summarise the articles.

Mem9 — Persistent Memory

Installing the Mem9 Skill

Message your Telegram bot to install the Mem9 skill using natural language:

> "Read https://mem9.ai/SKILL.md and follow the instructions to install and configure mem9 for OpenClaw"

OpenClaw will automatically fetch the skill instructions and guide you through the setup.

Configuring Mem9

After installation, OpenClaw will provide you with a MEM9 Space ID. Save it to Zeabur → Variable tab:

MEM9_SPACE_ID=your_mem9_space_id

The skill automatically registers itself as a memory plugin, replacing OpenClaw's default memory-core slot. Send /restart to your Telegram bot to activate.

How Memory Works

With Mem9 configured, your assistant remembers context across conversations. For example:

> "Remember that my favourite programming language is Python and I'm interested in agentic AI."

Then in a new session, ask:

> "What do you know about me?"

The agent will recall your preferences from the Mem9 store.

TiDB Cloud Zero — Database

Provisioning a Database

Message your Telegram bot to set up TiDB Cloud Zero using natural language:

> "Read https://zero.tidbcloud.com/SKILL.md and follow the instructions to create a database using TiDB Cloud Zero."

OpenClaw will provision a database and return the connection string.

Configuring the Database

Copy the connectionString from the agent's response. In Zeabur → Variable tab, add:

TIDB_CONNECTION_STRING=mysql://user:pass@gateway01.prod.aws.tidbcloud.com:4000/test

(Replace with your actual connection string.)

Send /restart to your Telegram bot to apply the changes.

Using the Database

Your agent can now use SQL to store and query data. For example:

> "Create a table to track the hackathon projects I've seen today, with columns for project name, team, and a short description."

TiDB Zero databases last 30 days and include a free quota plus a hard $1 spending limit — plenty for a hackathon demo. If you want to keep the database afterwards, you can claim it in TiDB Cloud.

QoderWork — Project Submission

When you're ready to present your work, QoderWork can build your submission deck automatically.

Preparing Your Submission

1. Screenshots — Capture key moments: your Telegram conversations, the OpenClaw dashboard with models configured, Bright Data search results, Mem9 memory recall, and any creative features you built
2. A short summary — 3–5 sentences describing what your agent does and which sponsor tools it uses

Generating Your Slide Deck

Open QoderWork and send a message like:

> "Create a hackathon submission presentation for the Agent Forge Hackathon. Here's my summary: [paste your summary]. And here are my screenshots: [attach your screenshots]."

QoderWork will generate a polished .pptx slide deck you can use directly for your hackathon submission, covering your project overview, architecture, feature demos with your screenshots, and sponsor acknowledgements.

Building Your Agent

With all five integrations live, your OpenClaw assistant can now search and scrape live web data (Bright Data), remember context across sessions (Mem9), store and query structured data (TiDB Cloud Zero), reason with a region-tuned LLM (Agnes AI via ZenMux), and communicate over Telegram.

Use Case Ideas

Here are some ideas to spark your imagination:

• Research Assistant — "Search for the top 10 AI startups in Singapore, summarise each one, and save them to a database so I can review later."
• Meeting Prep Bot — "Remember that I'm meeting John from Acme Corp tomorrow. Search for recent news about Acme and prepare 3 talking points."
• Competitor Tracker — "Every day, search for news about [competitor name] and store a summary in my database. Alert me if anything major happens."
• Learning Journal — "Remember everything I tell you about machine learning. When I ask for a quiz, test me on what I've learned."
• Event Scout — "Search for upcoming AI hackathons in Asia, filter by dates I'm available, and save the best ones."

The combination of web search, persistent memory, and a database means your agent can handle multi-step workflows that span days or weeks — not just single conversations.

Environment Variables Reference

Here's a quick-reference of all the environment variables you should have set:

# Zeabur & OpenClaw
TELEGRAM_BOT_TOKEN=...

# Agnes AI (via ZenMux)
ZENMUX_API_KEY=...

# Nosana
NOSANA_API_KEY=...

# Bright Data
BRIGHTDATA_API_KEY=...
BRIGHTDATA_SERP_ZONE=...

# Mem9
MEM9_SPACE_ID=...

# TiDB Cloud Zero
TIDB_CONNECTION_STRING=...

Troubleshooting

• OpenClaw won't start on Zeabur — Ensure you selected Dedicated Server (not Shared). OpenClaw requires at least 2 vCPU and 4 GB RAM.

• Telegram bot isn't responding — Double-check TELEGRAM_BOT_TOKEN in the Variable tab. After adding or changing it, restart the service. Then click "Connect Telegram Bot" in the dashboard.

• Bright Data skill not working — Verify both BRIGHTDATA_API_KEY and BRIGHTDATA_SERP_ZONE are set. Send /restart to your Telegram bot after adding them.

• Agnes AI model not appearing — Confirm the openclaw.json file was saved correctly with valid JSON syntax. Validate the configuration and restart the service.

• Mem9 not remembering across sessions — Check that MEM9_SPACE_ID is set correctly. Make sure the mem9 skill was installed successfully via natural language and send /restart to your bot.

• TiDB Cloud Zero connection failing — Verify the connection string format is correct (mysql://user:pass@host:4000/dbname). Zero instances expire after 30 days — ask your agent to provision a new one if yours has lapsed.

Quick Links

Summary

You've just built a production-ready AI assistant that can:

• Search the web in real time using Bright Data's SERP API
• Remember everything across sessions with Mem9's persistent memory
• Store and query data in a MySQL-compatible database via TiDB Cloud Zero
• Reason intelligently with Agnes AI and other powerful models
• Chat anywhere through Telegram

This is more than a demo — it's a foundation. Whether you're building a research assistant, a personal chief of staff, or something entirely new, OpenClaw gives you the building blocks to create agents that actually get things done.

We can't wait to see what you build. Happy hacking!