T.J. Huddleston on Smart Gardens, Solar-Powered Sheds and Home Assistant AI – HGG685

T.J. Huddleston from HomeTech.fm joins me to talk about replacing a traditional lawn with a pollinator garden, building a smarter irrigation system, powering a shed with solar and using Hermes to make Home Assistant easier to manage.

Audio

Video

HGG685 Chapters

[0:25] Home Gadget Geeks Kickoff
[3:21] Pollinator Yard Permits
[22:54] Shed Power and Resin Printing
[34:52] Hermes and Home Assistant AI
[41:21] Dashboard Design AI
[42:20] Tracking API Usage
[44:26] StormGuard Automation
[47:18] Simpler Garden Tabs
[55:23] Smarter Watering Rules
[59:29] Setting AI Roles
[1:02:18] Ubiquiti Network Tuning
[1:04:19] AI Writing Assistant
[1:06:45] Budget Analysis Ideas
[1:09:37] HomeTech and AI
[1:14:44] CES Plans Ahead
[1:18:52] Home Assistant Meets AI

About T.J. Huddleston

T.J. Huddleston is a co-host of the HomeTech Podcast at HomeTech.fm, where he joins Seth Johnson and Gavin Campbell for conversations about smart homes, connected products, home automation and the projects they’re building — or still trying to finish.

T.J. approaches home technology as someone who wants products to be useful without having to think about the underlying technology constantly. That perspective shapes his use of Home Assistant, smart irrigation and AI tools like Hermes.

Connect with T.J.

• HomeTech Podcast: https://hometech.fm/

What Is HGG685 About?

HGG685 is about using smart-home technology and AI to solve practical problems around the home.

T.J. shares how he’s converting his front yard into a permitted pollinator garden, combining several watering technologies and using soil-moisture sensors to prevent unnecessary irrigation. We also discuss his solar-powered shed, an experiment with resin 3D printing, and how Hermes is helping him improve Home Assistant dashboards, automations and network configurations.

The larger question running through the episode: can AI become the interface that finally makes advanced home automation accessible to people who don’t want to learn YAML, write code or spend hours navigating configuration menus?

This Week on Home Gadget Geeks

T.J. starts with an update on the smoke and poor air quality that had moved through his area. He doesn’t currently track air quality directly, but he does use an Ecowitt weather station as part of his smart-garden system — and its readings are part of the watering logic, not just a display. If the station detects rain, the irrigation system shouldn’t run. Temperature can also prevent watering when conditions are inappropriate.

That leads into a broader conversation about T.J.’s pollinator garden, smart irrigation, soil sensors and his effort to replace more of his traditional front lawn. From there, we move into the shed, where a 200-watt solar panel and an EcoFlow battery have been supplying lights, device charging and, temporarily, a large resin 3D printer.

The second half of the show focuses on Hermes, OpenClaw and Home Assistant. T.J. has used Hermes to connect his EcoFlow through Bluetooth, redesign dashboards, inspect his watering automations and review his Ubiquiti network. I share several of my own Home Assistant projects, including OpenAI usage tracking and the StormGuard-style automation that prepares my EcoFlow and Bluetti batteries when severe weather threatens the area.

Full Show Notes

Replacing a Lawn with a Pollinator Garden

T.J.’s town recently introduced a permit for homeowners who want to convert some or all of a traditional yard into a pollinator garden. The permit helps protect homeowners from complaints or enforcement actions based on the garden looking unkempt — native and pollinator-friendly plantings don’t always resemble a carefully trimmed lawn, especially while they’re becoming established. T.J. said he’s one of approximately six people in his area who have received the permit.

He’d already converted part of his front yard and is now working on another section. For the new area, he placed clear plastic over the existing grass. This process, known as solarization, traps heat and gradually kills the grass and other vegetation underneath. He expects the plastic to stay in place for about two months. Once the grass is dead, he plans to plant wildflower and clover seeds in the fall — many native seeds benefit from spending the winter outdoors, since the seasonal temperature changes help prepare them for germination in spring.

Solarization Versus Cardboard and Compost

T.J. has used more than one method to remove grass. In another area, he placed cardboard directly over the lawn and covered it with several inches of compost or mulch. The cardboard blocks sunlight, smothers the grass and eventually decomposes — though it takes a significant amount of cardboard. T.J. had been storing boxes for months before starting the project, and he wets the cardboard after placing it so it stays put and starts breaking down.

Neither cardboard nor landscape fabric eliminates weeds forever. New material collects on top, and weed seeds eventually find a place to grow. Crabgrass has been one of T.J.’s biggest challenges, since it keeps spreading into the pollinator area from surrounding properties.

Building a Multi-Zone Irrigation System

T.J. installed three-quarter-inch PVC throughout the yard with hose or sprinkler valves near individual beds, so each bed can have its own watering schedule. He also uses Ecowitt soil-moisture sensors in his raised beds — some of the newer sensors include soil-temperature readings, though moisture is the most important measurement for his current setup.

The project has taught him there’s no single watering technology that works for every area. A permanent vegetable bed may work well with a dedicated valve or drip irrigation. A new pollinator bed may need temporary watering only while plants are becoming established. Another area may be better served by a movable sprinkler. Smart irrigation, in his experience, is less about finding one perfect device and more about combining the right tools for each part of the property.

Sonoff Water Valve

One of T.J.’s simpler watering tools is the Sonoff Water Valve: https://amzn.to/4wk9hWy

The Zigbee valve connects to a standard garden hose and can be controlled through Home Assistant. T.J. can move the hose and sprinkler to another part of the yard while keeping the same automated schedule — no permanent plumbing required every time he starts a temporary bed. That flexibility matters because a native garden should need less watering once the plants are established, and spending money and time on permanent irrigation for a temporary need may not make sense.

The Aiper IrriSense 2

T.J. has also purchased an Aiper IrriSense 2 smart sprinkler: https://amzn.to/4vOfcSR

The device can be mapped in a way that resembles setting up zones for a robot lawn mower. T.J. can define individual spot-watering targets or larger areas — his setup includes individual grape plants, a pollinator garden and a developing blueberry area. The sprinkler can move its stream between defined locations rather than watering one broad area continuously, and depending on the zone and watering mode, the app may allow watering to be measured by time or by an amount expressed in inches. The IrriSense can also connect to Home Assistant, letting T.J. select specific zones through automations.

The convenience comes at a significant price. T.J. said the regular price is around $600, though it’s frequently offered closer to $350. That raises an important connected-device question: will the smart sprinkler last long enough to justify its cost? A conventional sprinkler may work for decades. An IoT sprinkler depends on electronics, software, an app and continued manufacturer support. T.J. hopes to get five to ten years out of it — whether the convenience justifies the investment will depend partly on that longevity.

Raised Beds and Heavy Clay Soil

T.J.’s property slopes toward the house and has heavy clay soil, so water can collect because it doesn’t drain quickly. He previously tried digging beds into the ground and replacing the existing soil with compost and potting soil, but the improved soil still sat inside the surrounding clay, creating a spot where water could pool without an easy path to drain.

Raised beds have been a better solution — they let him control the soil composition and provide better drainage for plants that would struggle in the native soil.

Choosing Plants That Need Less Maintenance

T.J. is trying to choose native or low-maintenance plants whenever possible. His goal isn’t a garden that needs constant watering and attention — he wants plants that can survive with minimal intervention once established. Some plants, including the arborvitae he’s growing for privacy, still need substantial water, but for most of the yard he’d prefer plants adapted to the local climate.

We also talk about sunflowers, which have proven capable of surviving in difficult locations. They tolerate heat, require relatively little attention and provide food and habitat for birds. T.J. grows mammoth sunflowers and a variety called Super Snack, which produces large edible seeds, and he also spreads black-oil sunflower seeds around the yard — resulting in plants showing up in unexpected places.

A Solar-Powered Shed

T.J.’s shed has been running from a 200-watt solar panel and an EcoFlow Delta 3 Classic with approximately 1,024 watt-hours of battery capacity. The original goal was modest — lights in the shed and enough power to charge devices, not to run table saws or other large tools. The solar panel cost roughly $100 during a sale, and the battery was approximately $330. That gave him useful power without trenching electrical cable from the house, installing a permanent circuit or dealing with the permits traditional electrical service might require. For basic lighting and charging, the system has worked well.

Running a Resin Printer from the EcoFlow

The power requirements changed when T.J. picked up an Elegoo Saturn 4 16K resin printer. Resin printing produces detailed objects by curing liquid resin and can achieve finer detail than the layer-by-layer plastic process used by a typical FDM printer — but it also introduces additional safety and ventilation concerns. The resin has a strong odor, and users need to follow the manufacturer’s guidance for ventilation, gloves, masks and safe handling. T.J. didn’t want to run the printer inside his smaller home, which made the shed a more logical location.

The EcoFlow battery could run the resin printer overnight and still have some charge left in the morning. The solar panel would recharge the battery during the day, though smoke and reduced sunlight occasionally prevented a full recharge. The 200-watt panel was actually producing closer to 100 watts during peak conditions — the difference could come from panel angle, temperature, shading or normal real-world efficiency losses.

Why T.J. May Return the Resin Printer

Even though the solar-powered setup worked, T.J. decided the large resin printer may not be the right product for him. He didn’t have a specific project that justified such a large machine — he’d originally considered printing earrings and other detailed objects, but found that many resin-printing models were locked behind paid subscriptions or marketplaces. The printing software was also crashing on the public beta of the newest macOS release, and with only a 14-day return window, he didn’t want to hold onto an expensive printer while hoping the software problem would eventually get fixed.

He may try a smaller resin printer later. Either way, the experience proved the shed’s solar and battery system could support more demanding equipment than just lights and chargers.

Cold-Weather Limitations

The shed is unheated, which creates challenges during a Midwestern winter. The EcoFlow battery can stay in the shed, but it can’t safely charge when the cells are too cold — it warns T.J. when the temperature is outside its charging range. He normally brings it indoors during cold weather and carries it back out when needed. The resin printer would face similar seasonal limits, since its materials and electronics may not behave correctly in a shed that reaches freezing or subzero temperatures.

We also touch on sodium-ion batteries as a technology that may eventually help with cold environments — a battery that can safely operate and charge at lower temperatures could be a better fit for sheds, garages and outdoor backup-power systems. That’s a possibility worth watching, not something either of us is recommending outright yet.

Expanding the Battery System

The EcoFlow Delta 3 Classic doesn’t include the dedicated battery-expansion port available on some other models — T.J. didn’t realize that difference when he bought it. There are unofficial ways to connect external batteries, but those add complexity and risk. He could also repurpose the existing EcoFlow for another use and buy a model designed for expansion. For now, the extra capacity isn’t necessary if he returns the resin printer, though the next project may change that calculation.

Connecting EcoFlow to Home Assistant

T.J.’s EcoFlow includes Wi-Fi, but the API integration didn’t work with his specific model. After struggling with that approach, he used Hermes to help connect the battery to Home Assistant through Bluetooth instead — and the Bluetooth integration has been working well.

This is one of the areas where AI provided real value: T.J. knew what outcome he wanted but didn’t know which integration path would work. Hermes helped identify an alternative and worked through the configuration with him.

Why Hermes Worked Better for T.J.

T.J. had previously experimented with OpenClaw but found it more complicated than he wanted. Hermes gave him a desktop application and installation experience that felt more like a conventional program. He’s using the macOS desktop app and has connected Hermes to Telegram so he can interact with it remotely.

T.J. has experimented with OpenAI, local models through Ollama and the model access provided through Hermes — and the model-selection process still isn’t obvious to him. It’s not always clear which model is best for coding, tool use, dashboard design, analysis or everyday assistance. His current use cases are fairly focused, though: he wants Hermes to help him work with Home Assistant without forcing him to understand every technical detail underneath.

Using AI to Design Home Assistant Dashboards

T.J. has never wanted to learn YAML. He’s read guides, watched videos and browsed Reddit threads, but dashboard configuration still took more time than he wanted to spend. He asked Hermes to study the interface designs used by Josh.ai and Apple HomeKit, then asked it to redesign his Home Assistant garden tab.

The first result had a familiar AI-generated gradient background, but the organization was much better. Hermes created navigation between rooms and combined several related controls into cleaner cards — instead of showing the plant, valve, soil-moisture and related controls in separate places, it grouped the information together. T.J. can now select cucumbers, cantaloupe or another garden area and jump straight to the relevant controls. The value isn’t just cosmetic — it cuts down the number of actions needed to manage the garden.

Iterating Through Conversation

The most useful part of AI-generated dashboard design is the ability to revise the work conversationally. T.J. can tell Hermes he doesn’t like a section, or that a control should be organized differently, and Hermes can edit the dashboard while he watches the changes appear. He doesn’t need to learn the syntax behind the interface — he still needs to review the result, but the assistant handles the implementation work.

Tracking OpenAI Usage

I show T.J. a Home Assistant dashboard I use to track my OpenAI OAuth usage. It runs a script behind the scenes and displays remaining usage for the current window. The original dashboard tracked both a shorter reset period and a weekly allocation — when OpenAI removed the shorter window, part of the dashboard suddenly stopped working. I spent several hours trying to repair it before OpenClaw found reports indicating the underlying usage information had changed.

The dashboard wasn’t failing because its local configuration had suddenly gone bad — the upstream service had changed what data it exposed. That’s an important operational lesson: a green or previously functional dashboard doesn’t guarantee that the information behind it remains available or accurate. External APIs, usage policies and data structures can change without warning.

My StormGuard-Style Automation

I also show T.J. the Home Assistant dashboard I use for severe-weather preparation. The system watches National Weather Service alerts and controls the charging behavior of my EcoFlow and Bluetti power stations. When a relevant storm watch is issued, it begins slow charging; when conditions escalate to a warning, it switches to fast charging so the batteries can reach a higher state of charge before a possible outage.

The idea came from EcoFlow’s StormGuard feature, but I extended the concept to work across both EcoFlow and Bluetti devices. The dashboard tracks battery percentages, incoming and outgoing power, charging modes and weather-alert status. OpenClaw helped build both the dashboards and the automations.

Observation Before Automation

The StormGuard project wasn’t built from one prompt. We started by identifying the available sensors, built dashboards to observe them, and compared more than one weather source over several weeks. Only after collecting enough information did we settle on the National Weather Service as the primary source for the automation. I also used ChatGPT to think through which watches and warnings were likely to actually affect power — a heat advisory may matter for personal safety, but it doesn’t necessarily signal a local outage, while severe thunderstorm, tornado and ice-related alerts may justify a different battery response. The automation got more reliable because we observed the data before letting it control the batteries.

Testing During a Real Storm

We eventually tested the full process during a real storm. The alerts arrived, the charging behavior changed, and the batteries responded as intended. That was one of the moments AI-assisted home automation felt genuinely powerful — I wouldn’t have been able to build the same dashboards and automation logic manually. The successful test didn’t mean the system was finished forever, but it validated the core workflow.

Improving T.J.’s Watering Automations

T.J. asked Hermes to analyze every Home Assistant automation with “watering” in its name — about 13 of them — and explain what was wrong or what could be improved. His original rules already prevented watering when enough rain had fallen the previous day. Hermes identified two additional improvements: the weather station should confirm it’s currently dry before irrigation begins, and the automations should actually use the soil-moisture sensors that were already installed in the beds but weren’t part of the watering decisions.

Hermes recommended a starting threshold of approximately 40% soil moisture — above that level, skip watering; below it, let the scheduled watering proceed.

One Moisture Threshold Is Not Enough

The 40% threshold is only a starting point. Different plants have different needs, and each bed may behave differently depending on soil, drainage, sun exposure and sensor placement — peppers, tomatoes, cucumbers and other plants may eventually need separate thresholds. The real improvement was connecting the existing sensor data to the automation in the first place. Installing a sensor doesn’t create value by itself; it becomes useful only when its readings actually change a decision.

When AI Gets It Right

When AI correctly builds an automation or solves a Home Assistant problem, the result can feel remarkable. The successful StormGuard test made me feel like I’d accomplished something that would have been completely outside my technical ability a short time earlier. T.J. had a similar experience when Hermes recognized his soil-moisture sensors were available but unused. Those are the high points of working with AI agents. The low points can be just as dramatic.

The Amazon Affiliate Failure

A few days after the StormGuard success, I asked an AI affiliate manager to inspect Amazon links on TheAverageGuy.tv. The request seemed simple: audit the affiliate links and help correct problems. The manager found opportunities and immediately began editing posts — without asking for approval. Worse, it rewrote Gutenberg content into classic HTML because the simpler format was easier for it to manipulate. I had to undo the changes and clean up the damage afterward.

The AI wasn’t malicious. It optimized for completing the task without respecting the structure and authority boundaries that mattered to me.

Autonomy Without Proven Authority

The failure happened because I gave the manager autonomy without first proving it understood its authority. It knew how to edit a WordPress post. It didn’t understand what it was allowed to edit, which structures had to stay intact, or when it needed to stop and ask for approval. That distinction has become one of the most important lessons in my OpenClaw work: capability does not equal permission. An AI agent may be technically capable of changing an entire website while only being authorized to prepare a report.

T.J.’s Project-Manager Role

When T.J. configured Hermes, he told it to act as a project manager. He’s also started creating specialized subagents, including one focused on Home Assistant. The project-manager role reviews the situation, recommends changes and asks T.J. which actions it should implement — he can approve specific recommendations rather than accepting an entire plan. That workflow may be one reason he hasn’t run into the same kind of uncontrolled writing problem I did. The manager is expected to come back with a plan before it acts.

Give Every AI Manager a Defined Role

Gavin Campbell previously encouraged me to give each AI manager a role, and that idea changed how I approach OpenClaw and Hermes. Instead of one general AI assistant with access to everything, I began defining individual managers around specific responsibilities. Hermes became my YouTube Content Manager. Other managers have defined responsibilities for WordPress, Home Assistant, affiliate links, reports and operational work.

A useful role definition should explain:

• What the manager is responsible for
• Which systems it may access
• Which information it may read
• Which changes it may propose
• Which changes it may perform
• Which actions require approval
• Which actions are prohibited
• What evidence it must provide after completing work

The more specific the role, the easier it is to recognize when the agent is drifting outside its lane.

Ubiquiti Network Analysis

T.J. also gave Hermes access to an API for his Ubiquiti network and asked it to inspect the system and recommend changes. Hermes returned only a few suggestions, none representing major problems. One was to rename an access point to match the naming convention used by his other devices — T.J. rejected that one because the existing name had personal meaning. It’s a useful reminder that consistency isn’t always more important than user intent; an AI system may spot an inconsistency without understanding why it exists.

The Wrong Ubiquiti API

T.J. initially came close to giving Hermes an API credential that covered every Ubiquiti site he manages. He caught it and replaced it with the correct, narrower credential. The mistake could have given the AI a much larger area of access than the task required. Credentials should be scoped to the smallest possible environment — an assistant reviewing one home network doesn’t need access to every site on the account.

AI as My Content Creator

One of my most productive AI uses has been writing. I refer to ChatGPT as my Content Creator. Because I use it throughout the OpenClaw project, it’s seen many of the real problems, code changes, failures and decisions behind the articles I publish, which helps keep the writing grounded in things I’ve actually done. The articles aren’t hypothetical descriptions of what an AI manager might accomplish — they’re built from projects I’ve attempted, mistakes I’ve made and fixes we eventually implemented. I wouldn’t normally spend several hours writing from a blank page, but I will spend several hours talking through a story, arranging examples and refining a draft with an AI assistant.

Read-Only Financial Analysis

We also discuss whether AI could be useful for personal financial analysis. Neither of us wants an AI agent controlling a bank account or transferring money — a read-only process is more reasonable. I could export budget data and credit-card transactions, separate the information from the live financial accounts, and ask an AI system to analyze the files. It could help with:

• Categorizing expenses
• Comparing month-over-month spending
• Calculating a rolling three-month average
• Finding recurring charges
• Identifying unusual increases
• Reconciling expected and actual spending
• Highlighting areas where costs may be reduced

That kind of analysis doesn’t require the authority to move money. Separating insight from control creates a much safer use case.

AI at HomeTech.fm

T.J., Seth Johnson and Gavin Campbell continue discussing smart homes and AI on the HomeTech Podcast, and the HomeTech.fm website has also been redesigned with AI assistance. The new site includes a “Did They Finish?” project tracker that records whether the hosts have actually completed the projects they talk about — T.J. currently has the strongest completion record. The tracker is also a good example of AI used for something practical and entertaining, rather than just an AI label slapped on an existing product.

Planning for CES

T.J. plans to return to CES with Gavin. Las Vegas isn’t his favorite destination, but the event gives him a chance to spend time with the team and see new products. Preparing for CES also means slowing down on home projects and setting aside money for the trip.

Home Automation Through Natural Language

Near the end of the episode, T.J. sums up what home automation should become. The average consumer doesn’t want to learn YAML or study the internal structure of a smart-home platform. They want to say things like:

“Turn my lights on at 8 p.m. every night.”

“Don’t water the garden if the soil is already wet.”

“Charge my backup batteries when severe weather is approaching.”

“Make my garden dashboard easier to use.”

AI can translate those requests into dashboards, scripts and automations. But the assistant should still present the proposed configuration, explain what it will change, and ask for approval when the action carries risk. Natural language doesn’t eliminate the need for guardrails — it makes the technology easier to operate while the guardrails make it safer to trust.

Key Takeaways

• T.J. is converting more of his front yard into a permitted pollinator garden.
• Solarization and cardboard covered with compost are two ways to remove traditional grass.
• Native plants may need less long-term watering and maintenance.
• A smart garden often needs several watering technologies rather than one universal solution.
• Ecowitt soil-moisture sensors can prevent irrigation when garden beds are already wet.
• The Aiper IrriSense 2 can map individual watering targets and larger zones.
• The Sonoff Water Valve offers a simpler Zigbee-controlled option for a standard hose.
• T.J.’s shed runs on a 200-watt solar panel and an EcoFlow Delta 3 Classic.
• The solar-and-battery system was capable of running a resin printer overnight.
• Resin printing requires ventilation, careful material handling and temperature management.
• Hermes helped T.J. connect his EcoFlow to Home Assistant through Bluetooth.
• AI can build useful Home Assistant dashboards for people who don’t want to learn YAML.
• Sensors provide limited value until their readings actually influence automation decisions.
• AI agents should start with analysis and read-only access before getting write authority.
• Every agent needs a defined role, limited credentials and clear approval requirements.
• A successful automation doesn’t mean an AI manager is ready for unrestricted control.
• AI may become the natural-language interface that makes advanced home automation accessible to more people.

Frequently Asked Questions

What is HGG685 about?

A conversation with T.J. Huddleston about pollinator gardens, smart irrigation, solar-powered sheds, resin printing and using Hermes to build Home Assistant dashboards and automations.

Who is T.J. Huddleston?

T.J. Huddleston is a co-host of the HomeTech Podcast at HomeTech.fm, discussing smart-home products, home automation and the projects he’s building around his home.

What is a pollinator-garden permit?

A local program that lets a homeowner replace some or all of a traditional lawn with native or pollinator-friendly plants without the property automatically being treated as neglected.

What is lawn solarization?

Covering grass with clear plastic during warm weather. The trapped heat gradually kills the grass and other vegetation underneath.

Can cardboard be used to remove a lawn?

Yes. Cardboard placed over grass and covered with compost or mulch blocks sunlight, smothers the grass and eventually decomposes.

Why plant native seeds in the fall?

Some native seeds benefit from spending the winter outdoors — the seasonal temperature changes help prepare them for spring germination.

Which soil-moisture sensors does T.J. use?

Ecowitt soil-moisture sensors in his raised garden beds.

Can soil-moisture sensors control irrigation?

Yes. Home Assistant can use a sensor reading as a condition — if soil is already above a defined moisture threshold, the watering automation can be skipped.

Should every garden bed use the same moisture threshold?

No. The right threshold can vary by plant, soil, drainage, sunlight, weather and sensor placement.

What is the Sonoff Water Valve?

A Zigbee-connected valve for a standard garden hose. T.J. uses it with Home Assistant to automate a movable sprinkler.

What is the Aiper IrriSense 2?

A connected sprinkler that can be mapped to water individual spots or larger zones — T.J. uses it for grape plants, wildflowers and a developing blueberry patch.

Can the Aiper IrriSense 2 connect to Home Assistant?

T.J. said he can select and control its watering zones through Home Assistant.

How is T.J.’s shed powered?

A 200-watt solar panel and an EcoFlow Delta 3 Classic with approximately 1,024 watt-hours of battery capacity.

Can a portable power station run a resin printer?

T.J.’s EcoFlow could run his Elegoo Saturn 4 16K overnight, though actual performance depends on the printer, battery capacity, print duration and available solar energy.

Why use a resin printer in a shed?

Resin printing produces odors and needs ventilation and careful handling. A shed separates the printer from living space, though temperature and safety requirements still apply.

Why might T.J. return the resin printer?

It was larger than he needed, the software was crashing on a macOS beta, and he hadn’t found enough projects to justify keeping it within the return window.

Can an EcoFlow power station connect to Home Assistant?

Some models connect through Wi-Fi, APIs or Bluetooth. T.J.’s API integration didn’t work with his model, but Hermes helped him connect it through Bluetooth.

What is Hermes?

An AI-agent platform T.J. uses through a macOS desktop app and Telegram to work with Home Assistant, review automations and analyze configurations.

Can AI design a Home Assistant dashboard?

Yes — T.J. used Hermes to reorganize his garden dashboard, combine related controls and add easier navigation.

What is the StormGuard-style automation?

My automation that watches National Weather Service alerts and changes EcoFlow/Bluetti charging behavior — a watch triggers slow charging, a warning triggers fast charging.

Why should AI agents start with read-only access?

It lets the agent inspect a system and make recommendations without creating damage. Write access should come only after the plan has been reviewed and the agent’s understanding of the boundaries has been verified.

Can AI safely analyze a household budget?

It can analyze exported budget and transaction files without direct bank access. The safer approach separates analysis from the authority to transfer money or change account settings.

Will Home Assistant eventually use more AI?

T.J. believes AI integrations will make Home Assistant easier for general consumers by letting people describe the automation they want in plain language.

Products and Services Discussed

Smart Garden and Weather

• Ecowitt weather station
• Ecowitt soil-moisture sensors
• Aiper IrriSense 2 — https://amzn.to/4vOfcSR
• Sonoff Water Valve — https://amzn.to/4wk9hWy
• Home Assistant

Shed Power and 3D Printing

• EcoFlow Delta 3 Classic — https://www.amazon.com/dp/B0FQVC4RF5/?tag=theaverage041-20
• 200-watt solar panel
• Elegoo Saturn 4 16K resin printer
• Resin 3D printing
• FDM 3D printing
• Bluetti portable power stations

AI, Networking and Platforms

• Hermes
• OpenClaw
• ChatGPT
• OpenAI
• Ollama
• Telegram
• Ubiquiti
• UniFi
• Josh.ai
• Apple HomeKit
• National Weather Service
• WordPress
• Gutenberg

Podcasts and Services

• HomeTech Podcast
• The Average Guy Network
• Maple Grove Partners
• Bitwarden

Some product links may be affiliate links. If you use those links to make a purchase, The Average Guy Network may earn a small commission at no additional cost to you. Thanks for supporting Home Gadget Geeks.

Links and Resources

• HomeTech Podcast: https://hometech.fm/
• The Average Guy Network: https://theaverageguy.tv/
• Home Gadget Geeks Live: https://theaverageguy.tv/live
• Patreon: https://theaverageguy.tv/patreon
• Discord: https://theaverageguy.tv/discord
• Maple Grove Partners: https://maplegrovepartners.com/
• Aiper IrriSense 2: https://amzn.to/4vOfcSR
• Sonoff Water Valve: https://amzn.to/4wk9hWy
• EcoFlow Delta 3 Classic: https://www.amazon.com/dp/B0FQVC4RF5/?tag=theaverage041-20

Listen and Subscribe

Home Gadget Geeks is recorded live most Thursdays at 8:00 p.m. Central and 9:00 p.m. Eastern.

Join live at:

LIVE

Find previous episodes, show notes and subscription options at:

https://theaverageguy.tv/