#a way to get assist to announce

1 messages · Page 1 of 1 (latest)

outer seal
#

is there a blueprint or way to get assist to announce what it sees from a image? I was looking at the options in a automation and I am still lost.

proper arrow
#

There's this integration that might be what you're looking for. I don't think you can pass an image to the text based Assist, even if it's backed by an multimodal LLM.

outer seal
proper arrow
#

Are you using the blueprint and this is ending with a notification action?

See developer tools --> actions --> llmvision.image_analyzer for direct access to the function that is most likely used inside the blueprint provided with the integration (haven't looked at it so far).

So you have to write a little script I guess, that calls this action and returns the LLMs response as the response_variable.

Expose that script to the LLM and give it a good description, then you should be able to ask it about things in an camera entitiy or image.

#

Ah, I see that you don't want to ask, you want to anounce it.
So you need an automation instead and trigger the action.
When you got the LLMs response, call assist_satellite.start_conversation as next step in the automation, or simply use TTS.

outer seal
#

yes just trying to get assist to announce the response from the image

#

thats the bit I am stuck on, getting the llm result to be announced via the assistant.

outer seal
#

so here is the weird thing, after the llm, I have this result

# 
Executed: August 6, 2025 at 10:03:56 AM
Error: expected str for dictionary value @ data['start_message']
Result:
params:
  domain: assist_satellite
  service: start_conversation
  service_data:
    preannounce: false
    extra_system_prompt: read out the response varible ai_analysis
    start_message:
      response_text: >
        The security camera image shows a residential street during daylight.
        Several parked cars are visible. No people or animals are apparent. No
        immediate security concerns are detected.
    entity_id:
      - assist_satellite.jarvis_voice_assist_satellite
  target:
    entity_id:
      - assist_satellite.jarvis_voice_assist_satellite
running_script: false
#

so its reading the message just not announcing it with the conversation

outer seal
#

it works via tts, but not via assist but includes the (response_text forward slashes in the announcement) lol

orchid marlin
#

Did you check out the blueprint using the new AI Task itegration menitoned in the release notes of 2025.8?

outer seal
outer seal
#

so here is what I get resault from the trying to use announce on voice assistant.

# 
Executed: August 9, 2025 at 7:30:30 AM
Error: expected str for dictionary value @ data['message']
Result:
params:
  domain: assist_satellite
  service: announce
  service_data:
    message:
      response_text: >-
        Here's a breakdown of what I observe in the image:


        **People:**


        *   There is one man visible. He is wearing a bright orange vest, a red
        shirt, black shorts, and white shoes. He is holding papers and what
        appears to be a small bag. He is walking along a sidewalk.


        **Vehicles:**


        *   A turquoise hatchback car is parked on the street.

        *   Two cars are parked in the parking lot in the background. One is
    preannounce: false
    device_id:
      - a77fb7400a25e2a3dfdd3243b6f5f294
  target:
    device_id:
      - a77fb7400a25e2a3dfdd3243b6f5f294
running_script: false```
#

this is the trace, notice the error and this is the code

# 
action: assist_satellite.announce
metadata: {}
data:
  message: '{{ ai_analysis }}'
  preannounce: false
target:
  device_id: a77fb7400a25e2a3dfdd3243b6f5f294
#

were as this works on tts

# 
action: tts.speak
metadata: {}
data:
  cache: true
  media_player_entity_id: media_player.esphome_web_0bac48
  message: "{{ ai_analysis }}"
  options:
    voice: en-gb-alan-low
target:
  entity_id: tts.piper