Introduction

This is a record of migrating the “make an Echo Dot talk from Home Assistant (HA)” setup through three generations. First alexa-remote-control driven over SSH, then HACS’s Alexa Media Player, finally the official Alexa Devices integration that landed in HA 2025.6. Around the same time I bulk-generated 75 automations with a Python script, then later consolidated 64 location notices down to 2 — a “proliferate then consolidate” operations phase. I also tried to build a custom skill calling ChatGPT from Alexa, and fell cleanly into an Amazon developer-account trap.

The “make an Echo Dot talk from HA” need was there from the start. Who entered or left which zone, did power consumption spike, is rain coming — I wanted those events spoken by the Echo Dot. But Amazon was long without an official “push TTS from local HA” API, so there’s a history of bolting on community hacks, and I ended up walking three generations of migration in order.

The three generations

Gen 1: alexa-remote-control (over SSH)

HA → SSH (port 2222) → alexa-remote-control → Amazon private API → Alexa speaks

Via shell_command.alexa_speak, SSH into the add-on and run alexa-remote-control inside. It works, but going over SSH adds latency, the shell-string escaping is fragile, and 2FA often jams it. (I’ve split the details of this generation into a separate post.)

Gen 2: Alexa Media Player (HACS)

HA → notify.alexa_media_echo_dot → Alexa speaks

Setup:

• Add alandtse/alexa_media_player as a custom repository in HACS
• Get the Amazon 2FA app key (52 chars) and paste it into the integration config
• Confirm the Echo Dot is recognized as media_player.echo_dot and that notify.alexa_media_echo_dot does TTS

Migrating the existing automations was simple: replace shell_command.alexa_speak with notify.alexa_media_echo_dot throughout automations.yaml (20 of them at the time).

Gen 3: Alexa Devices (official integration)

HA Core 2025.6 added the official Alexa Devices integration, which provides a notify.<device>_speak entity. The upsides of switching to the official one are big:

• Configuration is done entirely in the GUI (no HACS repo to add)
• You get a motion sensor (binary_sensor.<device>_motion) and an illuminance sensor (sensor.<device>_illuminance) — extras Alexa Media Player didn’t expose
• Maintenance rides on HA core, so it keeps up with Amazon-side spec changes faster

That said, notify.<x>_speak later went effectively deprecated on Amazon’s Alexa.Speak API — it returns 200 OK but is silent. You now have to switch to notify.<x>_announce (with a chime) — but that’s another post’s territory.

Bulk-generating 75 automations with Python

Right after switching to Gen 2, I bulk-generated automations with a Python script to have the Echo Dot announce every household event I could think of. The breakdown:

Location notices alone are 64. The approach emits one automation per “family member × zone” combination; that’s fine for the initial batch, but it means manufacturing more automations every time you add a zone. I consolidate that later.

Daily power-consumption sensor

To produce the 21:00 power report, I added a utility_meter to configuration.yaml:

utility_meter:
  daily_energy_consumption:
    source: sensor.smart_meter_total_consumption
    name: "Today's power use"
    cycle: daily

This generates a daily consumption sensor that resets at midnight. The 21:00 report covers: today’s energy (kWh) / current draw (W) / CO2 intensity (gCO2eq/kWh, via Electricity Maps) / fossil-fuel share (%).

To read this consumption data in the first place, you have to connect to the smart meter’s Route-B (the B-route) over Wi-SUN. I use RATOC’s Wi-SUN USB adapter (RS-WSUHA), which is quietly the priciest part of this whole setup — it cost over ¥20,000. And I’d let it sit idle for a while after buying it, with a low-key guilt of “this tiny USB-stick-looking thing that cost 20k is just gathering dust” (heh). This power report finally started earning it back.

Consolidating 64 location notices into 2

After running it a while, I folded the per-”person × zone” automations into 2.

• alexa_speak_person_enter_zone
• alexa_speak_person_leave_zone

Both are mode: parallel / max: 10, subscribing in one shot to the state triggers of each family member’s person.* entity. A Jinja template in the variables block fills in the person name and zone name dynamically, speaking it in one line.

The payoff of this pattern: adding a new zone requires no automation changes at all. Just add the zone to zones.yaml and enter/leave for the new zone is automatically spoken.

Not aiming for the consolidated form from the start was the right call — proliferating individually first, then consolidating once the zone data structure had settled, kept the abstraction on-target.

The Alexa-ChatGPT skill and the Amazon developer trap

I also built a custom skill that sends what you say to Alexa to the OpenAI API (GPT-4o-mini) for an answer (based on k4l1sh/alexa-gpt; Alexa-hosted Python; GptQueryIntent + AMAZON.SearchQuery slot — note {query} alone isn’t allowed in samples, a carrier phrase is required).

The thing that ate the most time here was the Amazon developer-account trap.

• developer.amazon.co.jp does not exist (DNS won’t resolve)
• Logging into developer.amazon.com with an amazon.co.jp account gives a password error
• Cause: if you use the same email for amazon.com and amazon.co.jp, the passwords conflict
• Fix: change the amazon.co.jp password so it differs from the amazon.com one

Not knowing this burns hours of “but the password is definitely right and it won’t go through.”

File ownership when touching HA from the SSH add-on

The SSH user (uid 1000) needs sudo to write under /homeassistant/ (root-owned, 644). The equivalent /config is a read-only mount you can’t write at all (the add-on sandbox); the real path is /homeassistant/.

The HA CLI (ha core check / ha core reload) errors out needing a token — ha’s auth isn’t injected into the SSH add-on shell. To reload automations from SSH you have to hit the HA Core REST API directly; mint a short-lived JWT from the jwt_key in .storage/auth and Bearer auth goes through without issuing a new token.

Wrap-up

• HA → Alexa speech reached the official integration through three generations of migration: alexa-remote-control → Alexa Media Player → Alexa Devices.
• Walking both a “mass-produce individual automations” phase and a “consolidate them with dynamic trigger + Jinja” phase is realistic (aim for consolidation from the start and the zone data structure never settles).
• If you’re building a ChatGPT skill on Alexa, clear the amazon.com / amazon.co.jp password conflict on the Amazon developer account before you start.

References

• Alexa Devices — Home Assistant
• alandtse/alexa_media_player (HACS)
• k4l1sh/alexa-gpt
• Utility Meter — Home Assistant
• Electricity Maps API