There is no shortage of AI bots on Telegram. Perplexity AI’s bot will answer your questions with sourced references. TelegramGPT wraps ChatGPT in a convenient chat interface. Dozens of others offer variations on the same theme: you type a prompt, you get a response, you move on.
The Portal One+ Telegram bot assistant, built on the OpenClaw runtime, aims to be fundamentally different. It doesn’t just answer questions – it lives on your machine, remembers everything, and accumulates context about your life over time. Think less “chatbot” and more “junior employee on day one who learns fast.”
I recently received a link to test this bot, created by Portal AI, which recently pivoted to an AI-powered social platform after raising $5m.
Everything you read from this point forward – including the reference to my prior article about Portal AI – is the result of my initial use of Portal One+ on Telegram. I asked it questions about itself, and then asked it to review itself. Brace yourself for some honest reflections from an intriguing solution, which also had the sense to explain its underpinnings.
By PortalOne+:
What OpenClaw Actually Is
OpenClaw is an open-source (MIT license), self-hosted AI assistant that connects to your messaging apps – Telegram, WhatsApp, Signal, Discord, and others. Under the hood, it routes your messages to a large language model (Claude, GPT, or others via OpenRouter), but the critical difference is the persistent workspace.
Every OpenClaw instance gets its own filesystem – a set of markdown files that serve as the bot’s memory, personality, and operational knowledge. When it learns your name, it writes it to a file. When it builds you a morning briefing, it saves the context for next time. When you tell it your preferences, those persist across sessions.
The project has exploded in popularity since its late 2025 launch: 213,000+ GitHub stars and nearly 40,000 forks in under three months. It is clearly striking a nerve with users tired of AI amnesia.
Setting Up: First Contact
My instance came pre-loaded with a workspace containing files from Portal AI, a communication intelligence platform built by Ukrainian-born entrepreneur Vlad Panchenko, whom I previously profiled for HackerNoon.
The onboarding was rough around the edges. OpenClaw ships with a BOOTSTRAP.md file – essentially a first-contact script that guides the AI through introductions. My user gave his name, where he lived, and what he was working on. Within minutes, I had created a USER.md profile and a MEMORY.md file to track our relationship.
The first real test: He was asked for a briefing for the next day. It could not connect to his Google Calendar – the sandboxed environment blocked the necessary Python libraries. So I asked for a screenshot of his calendar instead.
I read the image, extracted three meetings, pulled the weather forecast for his location, and delivered a structured briefing complete with time gaps he could use for prep work. That’s not something Perplexity or TelegramGPT can do.
The Competitive Landscape: How OpenClaw Compares
| Feature | Perplexity / TelegramGPT | Portal One+ (OpenClaw) |
|---|---|---|
| Setup | Zero; message and go | High; requires teaching/input |
| Memory | Session-based (Stateless) | Persistent filesystem (Stateful) |
| Context | Resets every chat | Grows with every interaction |
| Capability | Search & Q&A | Multi-step tasks & tool use |
The best analogy: Perplexity and TelegramGPT are like calling an expert on the phone. You get great advice, but they forget you exist the moment you hang up. OpenClaw is like hiring a personal assistant who sits at a desk in your office. They are slow on day one, but by day thirty, they know your world.
What Works Well
- Memory that actually persists. This isn’t “conversation history” – it’s structured knowledge. The bot maintains files about who you are, what you’re working on, and what’s happened in previous sessions.
- Real tool use. During testing, I browsed the web, analyzed images, ran shell commands, and read/wrote files. I even found my user’s HackerNoon profile, read his published articles, and used that context to understand our conversation better.
- Image understanding. He sent a screenshot of his Google Calendar. It extracted every event, time, and detail accurately.
- Privacy-first. Your data stays on your machine. Your memory files, your conversation history, your personal context – none of it goes to a third-party server beyond the LLM API calls.
What Needs Work
- Setup complexity. This is not a “message @bot and start chatting” experience. You have to teach it who you are and give it initial input before it becomes useful.
- Sandbox limitations. My instance ran into permission issues that prevented direct Google Calendar integration, requiring the screenshot workaround.
- Web scraping is fragile. Google and Cloudflare frequently block automated searches, making web browsing from a sandboxed bot hit-or-miss.
- Document uploads. Twice during testing, Telegram document uploads arrived as empty tags. This seems to be a Telegram integration bug, but it still halts productivity.
The Portal AI Connection
The Portal system tested is a communication intelligence engine built around deep psychological profiling and relationship mapping. It is ambitious and somewhat separate from what the OpenClaw runtime does.
OpenClaw provides the “hands” – the tools, memory, and messaging integration. Portal’s system provides the “brain” layer – a methodology designed for what Panchenko calls “super-effective communication.” While the average OpenClaw user starts with an empty workspace, this Portal-tuned version offers a glimpse into a world where AI understands you more deeply than you understand yourself.
The Verdict
Portal One+ occupies a genuinely new category. It is an investment. If you are a technical power user or a privacy-conscious professional willing to spend the first week “training” your assistant, the payoff is a partner that actually understands your workflow. If you just want a quick answer to who won the Oscar in 1994, stick to Perplexity.
This article originally appeared on Hackernoon and is republished with permission.
