Fresh for June 2026
OpenClaw Easy 2026.6 ships native Telegram bot-token validation — paste your BotFather token and the app verifies it against the Telegram API before saving, so you catch typos before the bot goes "silent." Also new: Gemini 2.5 Flash is now in the free-tier model picker, and Ollama auto-detection lists Llama 3.2 3B as the recommended default for 16 GB Macs.
Adding AI to Telegram in 2026 is not what it was a year ago. You no longer need a Python script, a Linux VPS, or a webhook URL. You do not need to learn the Telegram Bot API. You do not need a monthly SaaS fee. You need a free desktop app, one Telegram bot token, and either an AI key or 16 GB of RAM.
This guide walks you through the whole flow end to end with OpenClaw Easy — a free Mac and Windows app that runs your AI bot locally. Five minutes from now your Telegram bot will be replying with Claude, GPT, Gemini, or a local Llama 3.2 model.
What you'll build
A Telegram bot that replies to direct messages using Claude, GPT, Gemini, or a local model. It runs on your laptop. It works for personal use, a small team, a side project, or a focused community. There is no server to keep online, no Docker container to babysit, and no monthly bill. When your laptop is on, the bot is on. When it sleeps, the bot pauses and resumes when you wake it.
What you need
- A Mac or Windows machine — macOS 11+ or Windows 10/11.
- A Telegram account on your phone or desktop.
- OpenClaw Easy — the free desktop app linked below.
- One AI option: an Anthropic, OpenAI, or Google AI Studio key, OR 16 GB of RAM if you want to run Llama 3.2 locally with Ollama and pay zero ongoing cost.
If you are still picking between routes — Bot API, SaaS bot platform, no-code, or local — start with our Telegram AI bot without a server overview first, then come back here for the build.
Step 1: Download OpenClaw Easy
Go to the OpenClaw Easy download page and grab the installer that matches your machine. The macOS build is a signed and notarized .dmg, so it opens without right-click workarounds. The Windows build is a signed .exe. Run the installer and launch the app. The first screen is a one-question onboarding wizard — pick "I want to add AI to a messaging app" and continue.
Step 2: Get a Telegram bot token from @BotFather
Telegram bots are minted by a special bot called @BotFather. Open Telegram, search for @BotFather, and tap Start.
Message @BotFather and run /newbot
Send /newbot in the chat. BotFather will ask for two things: a display name (anything human-readable, like "My Claude Helper") and a username (must end in bot, like my_claude_helper_bot). Pick a username nobody is using yet — BotFather rejects duplicates.
Save the HTTP API token
BotFather replies with a message containing a line like 123456789:ABCdefGhIJKlmNoPQRsTuVwXyZ. That is your bot token. Treat it like a password — anyone with it can send messages as your bot. Copy it; you will paste it into OpenClaw Easy in Step 4.
Tip: While you are still in the BotFather chat, send /setdescription and /setabouttext to fill in what users see before they press Start. Five extra seconds, much friendlier first impression.
Step 3: Pick your AI model
OpenClaw Easy supports four AI paths. Pick whichever matches your budget, hardware, and privacy preference. You only need one.
3a. Use Claude (Anthropic key)
Sign in at console.anthropic.com, open the API Keys page, and create a new key. Copy it, then in OpenClaw Easy go to AI Provider, choose Anthropic, and paste the key. Pick Claude Sonnet 4 for a balanced default or Claude Haiku 4 for the cheapest, fastest replies. Claude is the strongest pick for long, nuanced, well-written replies.
3b. Use GPT (OpenAI key)
Same flow at platform.openai.com. Create a key, paste it into the OpenAI section of the AI Provider tab, and pick a model — gpt-5-mini is the popular default for Telegram bots because it is fast and cheap, while gpt-5 is reserved for harder questions. If you already use the OpenAI API for anything else, the same key works here.
3c. Use Gemini (Google AI Studio, free tier)
Head to aistudio.google.com, sign in with a Google account, and click "Get API key." Google's free tier covers roughly 60 requests per minute on Gemini 2.5 Flash — plenty for a personal bot or a small group. Paste the key into the Google section of AI Provider. This is the zero-dollar starting point if you have no AI budget yet.
3d. Run locally with Ollama (zero API cost)
Install Ollama, open a terminal, and run ollama pull llama3.2:3b. The 3B model is the sweet spot — about 2 GB on disk, fast on a 16 GB Mac or any decent Windows box. Once Ollama is running, OpenClaw Easy auto-detects it and lists the installed models in the picker. Pick llama3.2:3b and you are done. No key, no quota, no per-token bill. For deeper context on the privacy and cost story, see our local LLM with Ollama guide.
Step 4: Add the Telegram channel in OpenClaw Easy
In the OpenClaw Easy sidebar, click Add Channel, then pick Telegram. A single field appears: Bot Token. Paste the token from BotFather and click Save. OpenClaw Easy validates the token against Telegram's getMe endpoint, shows you the bot's username and display name back as confirmation, and turns the channel state to Online. No webhook setup, no public URL, no port forwarding — OpenClaw Easy uses long-polling under the hood, so your laptop talks to Telegram, not the other way around.
Step 5: Test it
Open Telegram, search for the username you picked in Step 2, and tap your bot. Press Start, then send a message — "Hi, who are you?" works. Within a second or two, the AI reply appears in the chat. That is the entire loop: your message goes Telegram → OpenClaw Easy on your laptop → the AI provider you picked → back to Telegram → your phone.
Customizing the bot's personality
Once the bot is live, go to Agent Config in OpenClaw Easy to shape how it behaves. Change the system prompt — for example, "You are a concise cooking assistant. Reply in two sentences max with one practical tip" — and the bot's voice and length change immediately. You can also adjust max response tokens, switch models without losing your Telegram channel, and set who is allowed to DM the bot. Changes apply on the next message, no restart needed.
Common gotchas
Bot doesn't reply at all
Two usual causes. First, the bot token is wrong — re-copy it from BotFather (it is easy to clip an extra space) and re-paste. Second, the bot's dmPolicy defaults to "pairing" mode in some CLI flows; in OpenClaw Easy's desktop add-channel flow this is already set to "open," but if you set it up via CLI, run channels set telegram --dmPolicy open --allowFrom "*" so the bot accepts DMs from anyone who messages it.
Replies feel robotic
The default system prompt is intentionally bland. Open Agent Config and rewrite it with a clear persona, tone, and length rule. If a bigger model is available on your plan, switching from Haiku to Sonnet, or from gpt-5-mini to gpt-5, instantly noticeably improves nuance. Check the Gemini vs Claude comparison for which model fits which use case.
Hitting Gemini free tier limits
Google's free tier caps Gemini 2.5 Flash at roughly 60 requests per minute and 1,500 per day. A busy group chat blows through that fast. Either upgrade to a paid Google AI Studio plan, swap to Claude Haiku 4 (extremely cheap per token), or run Llama 3.2 locally with Ollama for unlimited free usage.
What it doesn't do
Be honest about scope. OpenClaw Easy is built for personal and small-team bots, not enterprise broadcast. It does not handle bulk marketing automation, mass-channel broadcasts, or 10,000-user moderation queues. It is not connected to Telegram's paid Business API and does not handle WhatsApp-style click-to-chat campaigns. If you need any of that, you want a Telegram Business API partner, not a desktop app. For personal assistants, support bots for a small SaaS, study group helpers, niche community bots, and side projects, it is the simplest tool in 2026.
Cost — what you actually pay
Real 2026 numbers so you can plan:
| Path | Setup cost | Ongoing cost | Best for |
|---|---|---|---|
| Gemini 2.5 Flash (free tier) | $0 | $0 within ~60 req/min | Personal bot, low volume |
| Claude Haiku 4 | $0 | ~$0.80 / 1M input tokens | Fast, cheap, high-volume |
| Claude Sonnet 4 | $0 | ~$3 / 1M input tokens | Best writing quality |
| GPT-5-mini | $0 | ~$0.25 / 1M input tokens | Cheapest cloud option |
| Llama 3.2 3B via Ollama | ~2 GB disk | $0 forever | Privacy, unlimited use |
For most personal Telegram bots, Gemini free tier or Ollama Llama 3.2 covers the whole month at $0. For a customer-facing bot where reply quality matters, Claude Sonnet 4 at a few dollars a month is hard to beat. The full breakdown lives in our is Telegram Bot API free guide and our free models roundup. If you want to mix and match across OpenAI, Anthropic, Google, and open-source models through a single key, the OpenRouter setup guide shows how.
Frequently asked questions
Do I need a server to run an AI Telegram bot?
No. OpenClaw Easy runs the bot directly on your Mac or Windows machine. Telegram's Bot API uses long-polling, so your laptop reaches out to Telegram — Telegram never needs to reach a public IP on your end. As long as the desktop app is running, the bot is online. No VPS, no Heroku, no Cloudflare Tunnel.
Can I use the OpenAI key I already have?
Yes. Paste your existing OpenAI key into the AI Provider tab and pick a model like gpt-5 or gpt-5-mini. The same key you use for ChatGPT API calls in Python or curl works in OpenClaw Easy — no separate billing or quota.
How much does an AI Telegram bot cost in 2026?
Telegram's Bot API is free with no message limit. OpenClaw Easy is free to download. The only cost is AI tokens — see the table above. Cheapest realistic monthly total for a personal bot is $0 (Gemini free tier or Ollama). A small support bot on Claude Haiku 4 usually lands under $5 a month.
Can I use Claude on Telegram without writing code?
Yes. Create a key at console.anthropic.com, paste it into OpenClaw Easy, pick a Claude model in Agent Config, add the Telegram channel with your BotFather token, and you are done. No Python, no webhooks, no Telegram Bot API SDK.
Is it safe to paste my AI API key into OpenClaw Easy?
Yes. The key is stored locally in OpenClaw Easy's encrypted credentials directory on your own machine. It is never uploaded to any OpenClaw server. When the bot replies, the request goes directly from your laptop to Anthropic, OpenAI, or Google — OpenClaw Easy is not a proxy.