Setting up an autonomous personal AI assistant used to demand deep systems engineering and expensive cloud subscriptions. I reduced that entire setup to a straightforward, repeatable process by combining OpenClaw for orchestration, Ollama as my local engine, and Telegram as my mobile interface. My entire architecture runs on consumer hardware, requires zero cloud API fees, and lets me manage everything directly from my phone.
In the first step, I set up the local intelligence engine with Ollama. I install the runtime on my host machine to serve open weights locally, then pull an optimized model using the command: ollama pull llama3.2:8b. Once downloaded, I verify that the daemon is serving requests at
Next, I configure the bot endpoint through Telegram and BotFather. I message the official BotFather account, run the /newbot command, assign a display name, and pick a handle ending in _bot. BotFather then returns an HTTP API token, which functions as my remote bridge into the daemon.
For the third step, I handle orchestration with OpenClaw. I install the CLI and launch the onboarding process with the command: openclaw onboard. I point the engine to my Ollama instance at
Finally, I complete the pairing and security handshake. I send a simple message like "Hello" to my new bot in Telegram. My terminal detects the incoming connection and displays a security code. I approve my account with the command: openclaw pairing approve telegram PAIRING_CODE. With that authorization complete, unauthorized users are locked out, and the bot responds only to me.
The strength of this workflow lies in the clear division of labor. Ollama handles hardware acceleration and model weights locally. Telegram provides push notifications, voice notes, and cross-device access without exposing custom web servers to the public internet. OpenClaw ties everything together, turning simple chat messages into verified system actions on my machine.
