This ZeroClaw Telegram setup connects a bot to an existing working agent, approves your Telegram account, and checks the first model reply. The bot token connects ZeroClaw to Telegram; the pairing code authorizes a person. They are different secrets.
Tested October 14, 2026: ZeroClaw v0.8.5, Ollama 0.35.0, and qwen3:0.6b on this Mac. We used a temporary bot created on October 13, a fresh authorization configuration, channel alias home, and agent alias guide. The screenshots retain original browser pixels with crops, solid redactions, and arrows in added margins. The final reply succeeded after restarting the foreground listener; the earlier messages did not receive a reply.
Before starting, confirm your agent can answer a CLI request. For the local model configuration, see the Ollama and Telegram walkthrough. Run zeroclaw --version; these instructions target v0.8.5.
Step 1: Create the bot and connect your agent#
In Telegram, open the verified BotFather, send /newbot, and follow its display-name and username prompts. Copy the resulting bot token privately. See Telegram's official bot tutorial.

The temporary bot's token is hidden. Never copy a token from a tutorial; use your own bot.
On the ZeroClaw host, enter the token at the masked prompt:
zeroclaw config set channels.telegram.home.bot_token
zeroclaw config set channels.telegram.home.enabled true
zeroclaw agents list
zeroclaw config set agents.guide.channels
Replace guide with your existing agent alias. In the channel-list editor, add telegram.home while keeping any existing entries. The home alias is a local configuration name, not the bot's Telegram username. The agent must already have its model provider and risk profile configured.
The live test used the same channel and agent binding, with the token supplied through a private environment override rather than the recommended masked storage prompt. No token is included in this article or its images.
Step 2: Authorize your Telegram account#
For a new private bot, start with no authorized external peers for this channel. Do not clear access on an existing shared bot just to reproduce this test. Current access control uses peer groups, not an allowed_users field. Avoid the wildcard *: it opens access to everyone who can reach the bot and disables first-user pairing.
Start one foreground listener:
zeroclaw channel start
Keep that terminal open. For a fresh private configuration, it prints a one-time code. Open your bot's direct chat. Sending /start before authorization can produce an approval-required message; that is the expected access check. Send /bind followed by the code printed on your host, for example /bind YOUR_CODE with the placeholder replaced.

Live test: the account was initially unauthorized, then pairing succeeded. The numeric account ID and consumed pairing code are covered.
Successful pairing saves your numeric Telegram ID in the alias-scoped peer group. Our test retained authorization across the listener restart. If the bot reports that saving failed, fix that error before relying on persistence.
For an existing bot, an operator can authorize a known account on the ZeroClaw host:
zeroclaw channel bind-telegram YOUR_NUMERIC_ID --alias home
Replace the placeholder with the intended person's ID. If you are running a foreground listener, restart it after this separate CLI change. Consult the version-pinned Telegram authorization documentation for managed-service behavior and additional users.
Step 3: Verify an actual model reply#
Send a small request from the approved account:
What is 7 + 5? Answer in one short sentence.

The actual reply arrived at 10:34 on October 14: “7 + 5 equals 12.” This verifies delivery through Telegram, authorization, model inference, and the return message.
The test used Ollama at http://127.0.0.1:11434, model qwen3:0.6b, a 4,096-token context setting, thinking disabled, and an agent with tools disabled. Initial requests reached Ollama but produced no visible Telegram reply. We restarted the foreground listener and repeated the request; the reply above then arrived. The exact cause of the earlier delivery failure was not established. Pairing itself normally does not require a restart.
If your bot stays silent, check connectivity with zeroclaw channel doctor, verify the agent owns telegram.home, and inspect the listener output. Test the model independently. Run only one listener for the token; duplicate polling processes can conflict. After changing configuration, restart the relevant process and retry a short message.
Telegram still uses its cloud#
Local Ollama inference does not make this an entirely local chat system. Messages travel through Telegram's servers and Bot API to ZeroClaw, then back through Telegram. Bot conversations are not Telegram Secret Chats and are not end-to-end encrypted. A hosted model provider adds another external service; a local Ollama model removes that provider hop, not Telegram's cloud. See Telegram's bot FAQ and cloud versus Secret Chats.
Keep tokens and unused pairing codes out of shared screenshots and logs. If a bot token has been exposed, replace it through BotFather and update the host configuration.