Hi Dominik, sorry about that, and thanks for the screenshots. The store page was wrong, not your Homey.
The bridge only runs on Homey Pro (Early 2023), Homey Pro (2026), Homey Pro mini and Homey Self-Hosted Server. There is no setting to get around it on the 2019. You can safely uninstall it: it never started, so it left nothing behind.
Why not the 2019? The bridge talks to Homey through the newer local Homey API. A Homey Pro (Early 2019) offers the older version of that API, which is what the “platformVersion 1” in the message means. Devices, flows and settings come back in a different shape there, so every one of the 173 tools would need its own translation layer. The flow tools differ the most, and those are exactly the ones you were after. On top of that I don’t own a 2019, so I couldn’t test any of it properly. I’d rather not ship something I can only guess works.
The “Works on” list showed every model because the app only stated a firmware range, and your 2019 runs the same firmware numbers as the newer models. Version 4.20.3 fixes that: the store now only offers the app to the Homeys that can actually run it, so nobody else on a 2019 or older installs it for nothing.
For what you were after, on a supported Homey: the bridge runs on the Homey itself, so the cloud limits of Athom’s own MCP do not apply, and it can create an Advanced Flow straight into a folder (and create the folder first if needed). Sadly that doesn’t help on the 2019.
I’ve been testing the new Telegram chat feature in MCP AI Bridge. Really nice addition!
I’ve successfully connected a Telegram bot, but when using gpt-5.6-luna with Allow the model to act enabled, I receive the following error:
Function tools with reasoning_effort are not supported for gpt-5.6-luna in /v1/chat/completions. To use function tools, use /v1/responses or set reasoning_effort to ‘none’.
Would it be possible to add support for the OpenAI Responses API, allowing newer GPT models to use function calling and reasoning together?
I’d especially love to use GPT-6 with the Telegram integration to interact with and control my Homey devices.
Is this something you’re planning to support in a future update, or is there currently a workaround?
Thanks for your continued work on this app!
Also nice if the telegram bot can join my household group so all family members can ask things in the chat
@Mariomark1541 thanks for testing it, and for the exact error, that made it quick to find.
Newer OpenAI models such as gpt-5.6-luna reason by default, and OpenAI no longer allows reasoning and function calling together on the older chat completions endpoint, which the bridge still used for Ask a model with tools. 5.0.1 switches that to OpenAI’s Responses API, so tools and reasoning work together. Your Telegram bot, WhatsApp and the widget all go through the same path, so they all benefit.
Until 5.0.1 reaches the test channel, two workarounds: pick an OpenAI model that does not reason (gpt-4.1 or gpt-4.1-mini, for example), or switch the provider on the Ask a model tab to Anthropic, Gemini or Mistral.
Any newer OpenAI model that supports function calling in the Responses API will work the same way once you have 5.0.1, so give GPT-6 a try then and let me know how it goes.
@weide43 Thanks for the quick response and for fixing this in 5.0.1! Really appreciate it!
I’ll definitely give GPT-6 a try once the update is available and let you know how it goes. Tested and it’s working already with 5.6 Luna
I also have one more question about the Telegram integration.
Would it be possible to support Telegram group chats in the future? I’d love to add my AI bot to our family group so other family members can ask questions and interact with Homey as well.
Ideally, the bot would only respond when mentioned by its username (e.g. @BotNameBotName), so it doesn’t interfere with normal group conversations.
It would be great if group access could also be approved and managed separately for security reasons.
Is there a technical limitation preventing this, or could it be considered for a future update?
If you point the MCP AI Bridge at cloud models with your own keys, the daily bill is mostly prompt size: Claude Haiku 5.5 is $0.10 / $0.50 per million input/output under 100K prompts (cache reads $0.01/M; 5× above that band), Sonnet 5.5 is $2 / $10, and Grok 4.7 is $2 / $6 with $0.50/M cache reads — so a short “is the back door open?” turn stays under a cent, while pasting a week of power stats is what burns money. Cap max tokens on the Homey side, prefer Haiku or a cheaper Grok card for routine yes/no branches, and keep a prepaid balance so a runaway Flow cannot surprise you overnight. I keep a small public list-price table at https://aicost.tools/llm-cost/anthropic/claude-haiku-5-5/ (disclosure: I work on AICost) when I need to sanity-check a Flow before leaving it on.
Mention it once in the group, for example “@YourBotName hello”. The group then shows up on the Chat tab, marked as a group, waiting for approval.
Click Approve next to the group.
From then on, anyone in the group can ask it something by mentioning it or replying to one of its messages. The rest of the group conversation it leaves alone.
Each family member gets their own conversation with the bot, and the model is told who asked, so “turn off my lights” knows whose lights. /new@YourBotName starts a fresh conversation for just you.
Two things to keep in mind:
Everyone in an approved group, including people added later, acts under the same API key. Give the bot a key with Control only on the Security tab: it can switch things, but not rename, delete or change settings.
If the bot does not react to a mention, switch off its privacy mode at @BotFather (/setprivacy, pick your bot, Disable), then remove the bot from the group and add it again.
5.0.2 also has the Responses API change from 5.0.1, so GPT-6 should work there too. Let me know how it goes with the family!
@Mariomark1541 Safe enough for a family group, yes, if you set it up with a bit of care. What protects you:
Nothing reaches the model until you approve the group, and in the group the bot only reacts to messages that mention it or reply to it.
The bot acts under the API key you choose, through the same checks as Claude or ChatGPT: the key’s permission and tool selection, the access mode, your guardrails and the limits. Irreversible actions (deleting devices, uninstalling apps, rebooting) are refused for this route altogether.
Everything it does is logged on the Diagnostics tab, with the tool and the result.
The one thing to be aware of: anyone in an approved group can operate the house, including someone added to the group later. So:
Give the bot a key with Control only. It can switch and set devices, but not rename, delete or change your setup.
If there are things only you should control, such as a door lock or the alarm, block them with a guardrail on the Security tab. The bot then cannot touch them, whoever asks.
In the Telegram group settings, let only admins add members.
Set a daily question limit on the Ask a model tab, so a chatty afternoon cannot run up your OpenAI bill.
And it is a test version: if anything feels off, you can go back to the store version at any time. Let me know what you run into!
