A helpful assistant that runs on a Telegram interface. Started out as a branch from my homelabbing project, Pip emerged as a fun experiment for me to see what n8n is capable of with its AI-first workflow capabilities.
The system
At its core, n8n is a workflow tool just like any other — it listens for an input, does a process defined by the user, and ends with an output or closure. Pip isn’t any different; its input are either webhooks or listening for a Telegram message, the process an agentic AI system powered by tools, and the output a response to the webhooks or a reply on Telegram.
What’s fun are the tools that supercharge the agent’s abilities to do things. All of them are custom-made for the services and things I use, so they seem somewhat arbitrary but very personally useful. These tools include:
- Web search + knowledge gathering with SearXNG + Wikipedia tools
- Home lab management tools including Tailscale and command execution tools
- A Strava tool to tap into runs and other fitness activities (lol)
- A bookmarks tool to tap into saved links (Readeck)
- A calendar tool to fetch events from my calendars
- An LLM costs tool that aggregates the accrued costs from API calls
The model Pip uses is hot-swappable thanks to the Chat Model nodes n8n offers. Pip is currently using Anthropic’s Claude 4.5 Haiku — for some reason, I prefer the responses generated by this model than other competing LLMs.