Hey Whag,
A suggestion would be to look at this blueprint : Frigate Mobile App Notifications 2.0
Having the options this blueprint does and the buttons in the notification to silence new notifications or anything else you want would be awesome.
Thanks!
Hey this is super awesome since I was struggling with the original LLM Vision blueprint. However, I canât get a few things to work.
Here is my config
alias: Frigate LLM Notification V0.3 Custom
description: ""
use_blueprint:
path: Whag/frigate-review-with-llm-vision-notification-version-0-3.yaml
input:
in_camera: camera.doorbell
in_zones:
- 6c7f856a0bba564546cbd967127ed2e3
in_objects:
- person
- bicycle
- cat
- dog
- motorcycle
- car
in_host: http://192.168.4.133:8123
provider: 01JW4ZHMT4FYAWSWK7ZGHXCPVQ
model: qwen2.5vl:7b
in_gen_title: false
use_memory: false
use_remember: false
use_remember2: true
use_memory2: false
in_gen_title2: true
delay_clip: 2
delay_image: 2
in_notify_devices:
- f0847b4dab5e8f17e618e084e216c748
cooldown: 0
in_use_llm_image: false
in_lovelace_url: /dashboard-dashboard/front-door
Here is a recent trace
this:
entity_id: automation.frigate_llm_notification_v0_3_custom
state: 'on'
attributes:
id: '1748531469319'
last_triggered: '2025-05-29T20:25:08.516943+00:00'
mode: parallel
current: 0
max: 2
friendly_name: Frigate LLM Notification V0.3 Custom
last_changed: '2025-05-29T20:26:28.487748+00:00'
last_reported: '2025-05-29T20:26:28.487748+00:00'
last_updated: '2025-05-29T20:26:28.487748+00:00'
context:
id: 01JWEVK2R7SAHVG9ZN52ZZ97M0
parent_id: null
user_id: null
trigger:
id: End event
idx: '1'
alias: null
platform: mqtt
topic: frigate/reviews
payload: >-
{"type": "end", "before": {"id": "1748552094.598771-bwvofu", "camera":
"doorbell", "start_time": 1748552094.598771, "end_time": null, "severity":
"alert", "thumb_path":
"/media/frigate/clips/review/thumb-doorbell-1748552094.598771-bwvofu.webp",
"data": {"detections": ["1748551195.88762-v0hxth",
"1748552099.309217-kfop3r", "1748552092.987117-wazoiv",
"1748551194.932585-vgaa1w"], "objects": ["car"], "sub_labels": [], "zones":
[], "audio": []}}, "after": {"id": "1748552094.598771-bwvofu", "camera":
"doorbell", "start_time": 1748552094.598771, "end_time": 1748552134.303516,
"severity": "alert", "thumb_path":
"/media/frigate/clips/review/thumb-doorbell-1748552094.598771-bwvofu.webp",
"data": {"detections": ["1748551195.88762-v0hxth",
"1748552099.309217-kfop3r", "1748552092.987117-wazoiv",
"1748551194.932585-vgaa1w"], "objects": ["car"], "sub_labels": [], "zones":
[], "audio": []}}}
qos: 0
description: mqtt topic frigate/reviews
payload_json:
type: end
before:
id: 1748552094.598771-bwvofu
camera: doorbell
start_time: 1748552094.598771
end_time: null
severity: alert
thumb_path: /media/frigate/clips/review/thumb-doorbell-1748552094.598771-bwvofu.webp
data:
detections:
- 1748551195.88762-v0hxth
- 1748552099.309217-kfop3r
- 1748552092.987117-wazoiv
- 1748551194.932585-vgaa1w
objects:
- car
sub_labels: []
zones: []
audio: []
after:
id: 1748552094.598771-bwvofu
camera: doorbell
start_time: 1748552094.598771
end_time: 1748552134.303516
severity: alert
thumb_path: /media/frigate/clips/review/thumb-doorbell-1748552094.598771-bwvofu.webp
data:
detections:
- 1748551195.88762-v0hxth
- 1748552099.309217-kfop3r
- 1748552092.987117-wazoiv
- 1748551194.932585-vgaa1w
objects:
- car
sub_labels: []
zones: []
audio: []
frigate_topic: frigate/reviews
input_camera: camera.doorbell
input_objects:
- person
- bicycle
- cat
- dog
- motorcycle
- car
input_zones:
- 6c7f856a0bba564546cbd967127ed2e3
custom_zones: []
all_zones: false
host: http://192.168.4.133:8123
downloader: downloader
downloader_sub: null
notify_devices:
- f0847b4dab5e8f17e618e084e216c748
use_llm_image: false
gen_title: false
gen_title2: true
event:
type: end
before:
id: 1748552094.598771-bwvofu
camera: doorbell
start_time: 1748552094.598771
end_time: null
severity: alert
thumb_path: /media/frigate/clips/review/thumb-doorbell-1748552094.598771-bwvofu.webp
data:
detections:
- 1748551195.88762-v0hxth
- 1748552099.309217-kfop3r
- 1748552092.987117-wazoiv
- 1748551194.932585-vgaa1w
objects:
- car
sub_labels: []
zones: []
audio: []
after:
id: 1748552094.598771-bwvofu
camera: doorbell
start_time: 1748552094.598771
end_time: 1748552134.303516
severity: alert
thumb_path: /media/frigate/clips/review/thumb-doorbell-1748552094.598771-bwvofu.webp
data:
detections:
- 1748551195.88762-v0hxth
- 1748552099.309217-kfop3r
- 1748552092.987117-wazoiv
- 1748551194.932585-vgaa1w
objects:
- car
sub_labels: []
zones: []
audio: []
type: end
review_id: 1748552094.598771-bwvofu
input_camera_name:
- doorbell
camera: doorbell
camera_name: Doorbell
camera_match: true
detections:
- 1748551195.88762-v0hxth
- 1748552099.309217-kfop3r
- 1748552092.987117-wazoiv
- 1748551194.932585-vgaa1w
id: 1748551195.88762-v0hxth
objects:
- car
objects_match: true
zone_names:
- driveway
after_zones: []
zone_entered: false
all_zones_entered: true
zone_match: false
severity: true
notify_names:
- mobile_app_iphone
root_dir: downloader/
sub_dir: None/
video: /api/frigate/notifications/1748551195.88762-v0hxth/doorbell/clip.mp4
video_local: downloader/None/1748551195.88762-v0hxth_clip.mp4
thumb: /api/frigate/notifications/1748551195.88762-v0hxth/thumbnail.jpg
image: /api/frigate/notifications/1748551195.88762-v0hxth/snapshot.jpg
image_local: downloader/None/1748551195.88762-v0hxth_snapshot.jpg
gif: /api/frigate/notifications/1748552094.598771-bwvofu/review_preview.gif
frigate_msg: A car was detected in the
title_image: Doorbell - Alert
title_clip: Doorbell - Alert Update
And this is what my timeline looks like:
Amazing work with this though, excited to follow this!
Hi there, i can see there is no zones in the payload from the trace you supplied so that would explain this behaviour.
"objects": ["car"], "sub_labels": [], "zones":
[], "audio": []}}, "after": {"id": "1748552094.598771-bwvofu",
so if you had a zone input then you would not receive a notification, i would assume your frigate config needs altered to include âdrivewayâ in alerts.
My example below:
Regarding the timeline images i know this happens because i have not included the LLM Vision expose images which the timeline needs to show the image, if i add this you will be saving the image twice in different locations. I dont personally need this but iam happy to add this option in the next update/fix.
In my BP you also need to enable the ââUse Rememberââ input under images and clips if required.
Note: When configuring the timeline in LLM Vision you can set the auto delete days so you not storing these images in downlaoder,LLM and frigate for any longer than needed.
Hey thank you so much for the reply, I am fairly certain this will fix my issue with zones. As for the timeline, itâd be great to have that exposed even though its redundant, big fan of the LLM vision timeline.
Try this modified blueprint, not thoroughly tested tbh, I have got images in the timeline.
PUSHED TO LATEST!
blueprint:
name: Frigate LLM Notification Latest
author: whag
description: >
Basic requirements:<br />
- Home Assistant with an MQTT broker setup. <br />
- Frigate integration with MQTT setup. <br />
Link: https://github.com/blakeblackshear/frigate-hass-integration <br />
- LLM Vision integration with provider setup. <br />
Link: https://github.com/valentinfrlch/ha-llmvision <br />
- Downloader integration. <br />
Link: https://www.home-assistant.io/integrations/downloader/<br /><br />
Information:<br />
- Only suitable for one Frigate camera, use multiple instances of this automation for multiple cameras if required. <br />
- LLM Vision requires a provider setup for the automation to function, remember and memory options are optional. <br /><br />
Description: <br />
Use LLM Vision to analyse frigate snapshots and clips and send notifications via the Home Assistant companion app. <br />
Monitors MQTT frigate/reviews topic of types [new] and [end]. <br />
The Downloader integration has been used for storing the frigate files locally before sending them to LLM. In my experience this has improved the reliability.
Note that the downloader folder will need maintained separately (i.e. old files deleted or backed up). This can be achieved automatically via a shell command or similar <br />
This automation runs in parallel mode. <br /><br />
Tested on Android only. <br /><br />
Variables which may be useful for frigate message and AI prompts
- input_objects = Objects required<br />
- input_camera_name = Camera required<br />
- camera_name = Camera triggered<br />
- objects = Onjects found<br />
- zone_names = Zones required<br />
- after_zones = Zones triggered<br />
Version: v 0.4
domain: automation
input:
frigate:
name: Frigate Options
icon: mdi:cog
description: Frigate options
collapsed: true
input:
in_camera:
name: Frigate Camera
description: Frigate cameras only
selector:
entity:
multiple: false
filter:
- integration: frigate
domain: camera
in_zones:
name: Frigate Zones
description: Zones required for the alert. Filter will show all frigate devices(zones and cameras.)
default: []
selector:
device:
multiple: true
filter:
- integration: frigate
entity:
- domain: binary_sensor
in_custom_zones:
name: Custom zones
description: Add any custom zones not shown in the above selector.
default: []
selector:
text:
multiline: false
multiple: true
in_all_zones:
name: All Zones
description: All zones have to be entered before processing rather than any.
default: false
selector:
boolean: {}
in_objects:
name: Required Objects
description: Objects required in the selected zones if any.
default: []
selector:
select:
multiple: true
custom_value: true
options:
- person
- bicycle
- car
- motorcycle
- airplane
- bus
- train
- car
- boat
- traffic light
- fire hydrant
- street sign
- stop sign
- parking meter
- bench
- bird
- cat
- dog
- horse
- sheep
- cow
- elephant
- bear
- zebra
- giraffe
- hat
- backpack
- umbrella
- shoe
- eye glasses
- handbag
- tie
- suitcase
- frisbee
- skis
- snowboard
- sports ball
- kite
- baseball bat
- baseball glove
- skateboard
- surfboard
- tennis racket
- bottle
- plate
- wine glass
- cup
- fork
- knife
- spoon
- bowl
- banana
- apple
- sandwich
- orange
- broccoli
in_host:
name: Home Assistant IP Address.
description: Home Assistant ip address.
default: http://192.168.1.1:8123
selector:
text:
multiline: false
multiple: false
llm:
name: LLM Vision Provider / Model
icon: mdi:cog
description: LLM Vision options
collapsed: true
input:
provider:
name: Provider
description: Provider to use for analysis.
selector:
config_entry:
integration: llmvision
model:
name: AI Model
description: AI Model
default: gemini-2.0-flash
selector:
select:
options:
- gpt-4o
- gpt-4o-mini
- gpt-4.1
- claude-3-7-sonnet-latest
- claude-3-5-sonnet-latest
- claude-3-5-haiku-latest
- claude-sonnet-4-20250514
- claude-opus-4-20250514
- gemini-1.5-flash
- gemini-1.5-pro
- gemini-2.0-flash
- gemini-2.0-flash-lite
- gemini-2.5-pro-preview-05-06
- gemini-2.5-flash-preview-05-20
- llama-3.2-90b-vision-preview
- llama-3.2-11b-vision-preview
- Llama-4-Scout-17B-16E-Instruct-FP8
- Llama-4-Maverick-17B-128E-Instruct-FP8
multiple: false
custom_value: true
image_analyse:
name: Snapshot Analyse
icon: mdi:cog
description: Frigate event snapshot analyse options
collapsed: true
input:
in_title_image:
name: No AI Notification Title
description: Notification title for snapshot. If you are using LLM for image analyse and want a AI generated title ensure [Generate Title] below is enabled.
default: "{{camera_name}} - Alert"
selector:
text:
in_frigate_msg:
name: No AI Notification Message
description: Inital notificaion message if not using LLM Vision for the snapshot analysis.
default: "A {{objects| join(' and a ')|replace('_', ' ')}} was detected in the {{after_zones| join(' and the ')|replace('_', ' ' )}}"
selector:
text:
multiline: true
multiple: false
in_use_llm_image:
name: LLM Vision Snapshot
description: Use LLM Vision for intial snapshot analysis. If off then the custom message above will be used to generate the notification.
default: true
selector:
boolean: {}
prompt:
name: Prompt
description: Model prompt.
default: Summarise the image in a 1 sentence response,
concentrate on the {{input_objects| join(',')|replace('_', ' ')}} if seen within
the {{zone_names| join(' and ,')|replace('_',' ')}}.
Any person approaching the camera should take priority over all other events.
The response is an intial alert for a mobile phone security notification. If you see
a known person from memory, mention this so the response.
selector:
text:
multiline: true
multiple: false
target_width:
name: Pixels
description: Width in pixels to downscale.
default: 1080
selector:
number:
min: 512
max: 1920
step: 10
mode: slider
tokens:
name: Tokens
description: Maximum number of tokens to generate.
default: 100
selector:
number:
min: 1
max: 300
step: 1
mode: slider
temperature:
name: Response Creativity
description: Lower is more accurate, higher is more creative.
default: 0.5
selector:
number:
min: 0.0
max: 1.0
step: 0.1
mode: slider
in_gen_title:
name: Generate Title
description: AI generated notification title.
default: false
selector:
boolean: {}
use_memory:
name: Use Memory
description: Use information stored in memory to provide additional context.
Memory must be setup in LLM Vision.
default: false
selector:
boolean: {}
use_remember:
name: Use Remember
description: Store this event in the timeline.
Timeline must be setup in LLM Vision.
default: false
selector:
boolean: {}
expose_image:
name: Expose Image
description: Save image in llmvision for Timeline.
Timeline must be setup in LLM Vision.
default: false
selector:
boolean: {}
delay_image:
name: Delay Before Image
description: Delay in seconds before processing Frigate snapshot.
default: 2
selector:
number:
min: 0
max: 60
step: 1
mode: slider
clip_analyse:
name: Review Clip Analyse
icon: mdi:cog
collapsed: true
description: Frigate review clip analyse options
input:
in_title_clip:
name: No AI Notification Title
description: Notification title for review clip. If you want a AI generated title ensure [Generate Title] below is enabled.
default: "{{camera_name}} - Alert Update"
selector:
text:
multiline: false
multiple: false
prompt2:
name: Prompt
description: Model prompt.
default: You are viewing a clip from my {{camera_name}} security camera.You need
to concentrate on the {{zone_names| join(' and ,')|replace('_',
' ' )}} areas rather than other areas. The primary objects
of interest are {{input_objects| join(',')|replace('_', ' '
)}} describe them and what they are doing. If it is a
known {{input_objects| join(',')|replace('_', ' ' )}} or a
familiar company for example Royal Mail or DHL then state
this in your description. Try and keep the response direct
and clear and keep the response to 2 sentences so it can
be easily viewed on a mobile phone notification. Do not
reference this prompt, the weather or the environment in
the response.
selector:
text:
multiline: true
multiple: false
target_width2:
name: Pixels
description: Width in pixels to downscale.
default: 1080
selector:
number:
min: 512
max: 1920
step: 10
mode: slider
tokens2:
name: Tokens
description: Maximum number of tokens to generate.
default: 100
selector:
number:
min: 1
max: 300
step: 1
mode: slider
temperature2:
name: Response Creativity
description: Lower is more accurate, higher is more creative.
default: 0.5
selector:
number:
min: 0.0
max: 1.0
step: 0.1
mode: slider
max_frames:
name: Max Frames
description: How many frames to analyse for clip.
default: 3
selector:
number:
min: 1
max: 10
step: 1
mode: slider
in_gen_title2:
name: Generate Title
description: AI generated clip title.
default: false
selector:
boolean: {}
use_memory2:
name: Use Memory
description: Use information stored in memory to provide additional context.
Memory must be setup in LLM Vision.
default: false
selector:
boolean: {}
use_remember2:
name: Use Remember
description: Store this event in the timeline.
Timeline must be setup in LLM Vision.
default: false
selector:
boolean: {}
expose_clip:
name: Expose Clip
description: Save clip in llmvision for Timeline.
Timeline must be setup in LLM Vision.
default: false
selector:
boolean: {}
delay_clip:
name: Delay Before Clip
description: Time in seconds before processing the Frigate clip.
default: 2
selector:
number:
min: 0
max: 60
step: 1
mode: slider
Downloader:
name: Downloader Options
icon: mdi:cog
collapsed: true
input:
in_downloader:
name: Downloader Directory
description: Downloader integration root directory.
default: downloader
selector:
text:
multiline: false
multiple: false
in_downloader_sub:
name: Downloader Sub Directory
description: Define a sub directory within your downloader root directory.
default:
selector:
text:
multiline: false
multiple: false
download_image_wait:
name: Image Download Max Wait Time
description: Maximum time to wait in seconds for the frigate image to be downloaded before retrying download.
default: 5
selector:
number:
min: 1
max: 120
step: 1
mode: slider
download_clip_wait:
name: Clip Download Max Wait Time
description: Maximum time to wait in seconds for the frigate clip to be downloaded before retrying download.
default: 30
selector:
number:
min: 1
max: 120
step: 1
mode: slider
notify:
name: Notification Options
icon: mdi:bell
collapsed: true
input:
in_notify_devices:
name: Notify Devices
description: The devices to send the notification to. Multiple devices may be
used. Only works with Home Assistant Companion mobile app.
default: []
selector:
device:
multiple: true
filter:
- integration: mobile_app
click_action1:
name: Click Action New Event
description: Notification Click Action for New Event Notification . Default used is a Lovelace URL.
default: /lovelace
selector:
select:
options:
- label: View Downloaded Snapshot
value: "{{image_local}}"
- label: View Frigate Snapshot
value: "{{image}}"
click_action2:
name: Click Action End Event
description: Notification Click Action For End Event Notification . Default used is a Lovelace URL.
default: /lovelace
selector:
select:
options:
- label: View Downloaded Snapshot
value: "{{image_local}}"
- label: View Frigate Snapshot
value: "{{image}}"
- label: View Downloaded Clip
value: "{{video_local}}"
- label: View Frigate Clip
value: "{{video}}"
cooldown:
name: Cooldown
description: Time in minutes to wait before running this automation again.
default: 2
selector:
number:
min: 0
max: 60
step: 1
mode: slider
Actions:
name: Custom Actions
icon: mdi:bell
collapsed: true
input:
in_action1:
name: New Event Custom Actions
description: Custom action to run on new event.
default: []
selector:
action: {}
in_action2:
name: End Event Custom Actions
description: Custom action to run on end event.
default: []
selector:
action: {}
triggers:
- trigger: mqtt
topic: "{{frigate_topic}}"
payload: new
value_template: "{{value_json['type']}}"
id: New event
- trigger: mqtt
topic: "{{frigate_topic}}"
payload: end
value_template: "{{value_json['type']}}"
id: End event
conditions:
- condition: template
value_template: "{{camera_match}}"
alias: Valid Camera
- condition: template
value_template: "{{objects_match}}"
alias: Valid Object
- condition: template
value_template: "{{zone_match}}"
alias: Valid Zone
- condition: template
value_template: "{{severity}}"
alias: Valid Alert
actions:
- action: logbook.log
metadata: {}
data:
name: LLM Vision Frigate Notification log
entity_id: "{{this.entity_id}}"
message: >-
Event type: {{type}} /n
Review id: {{review_id}} /n
Input camera: {{input_camera}} /n
Camera triggered: {{input_camera_name}} /n
Detections ID's: {{detections}} /n
Detection 1 ID: {{id}} /n
Objects required: {{input_objects}} /n
Objects found: {{objects}} /n
Object match: {{objects_match}} /n
Zones required: {{zone_names}} /n
Zones triggered: {{after_zones}} /n
Zone match: {{zone_match}} /n
Notify devices: {{notify_names}} /n
Downloader root: {{root_dir}} /n
Downloader sub: {{sub_dir}} /n
Thumb url {{thumb}} /n
Snapshot url {{image}} /n
Clip url {{video}} /n
GIF url {{gif}}
- alias: Event is new or update Do LLM image analysis and send notification
if:
- condition: trigger
id:
- New event
then:
- alias: Run Custom Actions New
choose:
- conditions: "{{action1|length != 0}}"
sequence: !input in_action1
- delay:
hours: 0
minutes: 0
seconds: 2
milliseconds: 0
- alias: Repeat Until file exists or 2 retries
repeat:
sequence:
- action: downloader.download_file
continue_on_error: true
metadata: {}
data:
overwrite: true
url: "{{host}}{{image}}"
subdir: "{{sub_dir}}"
filename: "{{id}}_snapshot.jpg"
- alias: Wait for downloader to complete
wait_for_trigger:
- trigger: event
event_type: downloader_download_completed
event_data:
url: "{{host}}{{image}}"
filename: "{{id}}_snapshot.jpg"
id: image_exists
continue_on_timeout: true
timeout:
hours: 0
minutes: 0
seconds: !input download_image_wait
milliseconds: 0
- alias: If file exists run LLM Image Analyzer and send notification
if:
- condition: template
value_template: "{{ wait.trigger.id == 'image_exists' }}"
then:
- delay:
hours: 0
minutes: 0
seconds: !input delay_image
milliseconds: 0
- if:
- condition: template
value_template: "{{use_llm_image}}"
then:
- action: llmvision.image_analyzer
data:
provider: !input provider
model: !input model
remember: !input use_remember
expose_images: !input expose_image
use_memory: !input use_memory
image_file: "{{image_local}}"
message: !input prompt
include_filename: false
target_width: !input target_width
max_tokens: !input tokens
temperature: !input temperature
generate_title: "{{gen_title}}"
response_variable: response
- delay:
hours: 0
minutes: 0
seconds: 5
milliseconds: 0
- variables:
title: "{{response.response_title if gen_title else title_image}}"
msg: "{{response.response_text}}"
else:
- variables:
title: "{{title_image}}"
msg: "{{frigate_msg}}"
- alias: Send Intial notification to notify devices
repeat:
for_each: "{{notify_names}}"
sequence:
- action: notify.{{ repeat.item }}
data:
title: "{{title}}"
message: "{{msg}}"
data:
image: "{{thumb}}"
entity_id: "{{input_camera}}"
url: !input click_action1
clickAction: !input click_action1
group: "{{camera}}-frigate-notification"
tag: "{{id}}"
ttl: 0
priority: high
- delay:
hours: 0
minutes: !input cooldown
seconds: 0
milliseconds: 0
alias: Cooldown
until:
- condition: template
value_template: "{{wait.trigger.id == 'image_exists' or repeat.index == 2}}"
enabled: true
- alias: Event is end Do LLM video analysis and send notification
if:
- condition: trigger
id:
- End event
then:
- alias: Run Custom Actions End
choose:
- conditions: "{{action2|length != 0}}"
sequence: !input in_action2
- delay:
hours: 0
minutes: 0
seconds: 5
milliseconds: 0
- repeat:
sequence:
- action: downloader.download_file
continue_on_error: true
metadata: {}
data:
overwrite: true
url: "{{host}}{{video}}"
subdir: "{{sub_dir}}"
filename: "{{id}}_clip.mp4"
- alias: Wait for downloader to complete
wait_for_trigger:
- trigger: event
event_type: downloader_download_completed
event_data:
url: "{{host}}{{video}}"
filename: "{{id}}_clip.mp4"
id: clip_exists
continue_on_timeout: true
timeout:
hours: 0
minutes: 0
seconds: !input download_clip_wait
milliseconds: 0
- delay:
hours: 0
minutes: 0
seconds: !input delay_clip
milliseconds: 0
- if:
- condition: template
value_template: "{{wait.trigger.id == 'clip_exists'}}"
then:
- action: llmvision.video_analyzer
metadata: {}
data:
provider: !input provider
model: !input model
remember: !input use_remember2
expose_images: !input expose_clip
use_memory: !input use_memory2
video_file: "{{video_local}}"
max_frames: !input max_frames
include_filename: false
target_width: !input target_width2
max_tokens: !input tokens2
temperature: !input temperature2
generate_title: "{{gen_title2}}"
message: !input prompt2
response_variable: response
- delay:
hours: 0
minutes: 0
seconds: 1
milliseconds: 0
- alias: Update notification to notify devices
repeat:
for_each: "{{notify_names}}"
sequence:
- action: notify.{{ repeat.item }}
data:
title: "{{response.title if gen_title2 else title_clip}}"
message: "{{response.response_text}}"
data:
video: "{{gif}}"
image: "{{gif}}"
entity_id: "{{input_camera}}"
url: !input click_action2
clickAction: !input click_action2
group: "{{camera}}-frigate-notification"
tag: "{{id}}"
interruption-level: active
ttl: 0
priority: high
- delay:
hours: 0
minutes: !input cooldown
seconds: 0
milliseconds: 0
alias: Cooldown
until:
- condition: template
value_template: "{{wait.trigger.id == 'clip_exists' or repeat.index == 2}}"
trigger_variables:
frigate_topic: frigate/reviews
variables:
frigate_topic: frigate/reviews
input_camera: !input in_camera
input_objects: !input in_objects
input_zones: !input in_zones
custom_zones: !input in_custom_zones
all_zones: !input in_all_zones
host: !input in_host
downloader: !input in_downloader
downloader_sub: !input in_downloader_sub
notify_devices: !input in_notify_devices
use_llm_image: !input in_use_llm_image
gen_title: !input in_gen_title
gen_title2: !input in_gen_title2
event: "{{ trigger.payload_json }}"
type: "{{event['type']}}"
review_id: "{{event['after']['id']}}"
input_camera_name: >-
{{input_camera|expand|map(attribute='attributes.camera_name',
default='none')|list}}
camera: "{{event['after']['camera']}}"
camera_name: "{{ camera | replace('_', ' ') | title }}"
camera_match: "{{input_camera_name|select('equalto', camera)|list|length>0 }}"
detections: "{{ event['after']['data']['detections'] }}"
id: "{{ detections[0] }}"
objects: "{{ event['after']['data']['objects']}}"
objects_match: >-
{{input_objects | length == 0 or (input_objects |select('in', objects) |list
|length > 0) }}
zone_names: >-
{% set device_zones = input_zones %}
{% set zones = device_zones + custom_zones %}
{% set ns = namespace(device_names=[]) %}
{% for device_id in zones %}
{% if device_attr(device_id, "name") != None %}
{% set device_name = device_attr(device_id, "name") %}
{% set sanitized_name = device_name | lower | regex_replace("[^a-z0-9_\- ]", "") | replace(" ", "_") | replace("-", "_") %}
{% set ns.device_names = ns.device_names + [sanitized_name] %}
{% else%}
{% set ns.device_names = ns.device_names + [device_id | lower | regex_replace("[^a-z0-9_\- ]", "") | replace(" ", "_") | replace("-", "_")] %}
{% endif %}
{% endfor %}
{{ ns.device_names }}
after_zones: "{{ event['after']['data']['zones'] | default([]) }}"
zone_entered: >-
{{ zone_names | length == 0 or ((zone_names | select('in', after_zones) | list |
length) > 0) }}
all_zones_entered: >-
{{true if (zone_names | reject('in',after_zones) | list | length == 0 and all_zones) or all_zones == false else false }}
zone_match: "{{zone_entered and all_zones_entered}}"
severity: "{{event['after']['severity'] == 'alert'}}"
notify_names: >-
{% set ns = namespace(device_names=[]) %}
{% for device_id in notify_devices%}
{% set device_name = device_attr(device_id, "name") %}
{% set sanitized_name = "mobile_app_" + device_name | lower | regex_replace("[^a-z0-9_\- ]", "") | replace(" ", "_") | replace("-", "_") %}
{% set ns.device_names = ns.device_names + [sanitized_name] %}
{% endfor %}
{{ ns.device_names }}
root_dir: >-
{% set dir = downloader|trim %}
{%if dir|length > 0%}
{% if dir[-1] != "/" %}
{% set dir = dir + "/" %}
{% endif%}
{% endif%}
{{dir}}
sub_dir: >-
{% set dir = downloader_sub|trim %}
{%if dir|length > 0%}
{% if dir[-1] != "/" %}
{% set dir = dir + "/" %}
{% endif%}
{% endif%}
{{dir}}
video: /api/frigate/notifications/{{id}}/{{camera}}/clip.mp4
video_local: "{{root_dir}}{{sub_dir}}{{id}}_clip.mp4"
thumb: /api/frigate/notifications/{{id}}/thumbnail.jpg
image: /api/frigate/notifications/{{id}}/snapshot.jpg
image_local: "{{root_dir}}{{sub_dir}}{{id}}_snapshot.jpg"
gif: /api/frigate/notifications/{{review_id}}/review_preview.gif
frigate_msg: !input in_frigate_msg
title_image: !input in_title_image
title_clip: !input in_title_clip
action1: !input in_action1
action2: !input in_action2
mode: parallel
max: 2
Thanks! testing now
Love this blueprint so much, is there any way to have it so when I click the notification itâll take me to the saved image and caption? Sounds like a bigger task that it may be worth, but just wondered if thereâs an easy workaround?
EDIT:
For anyone else wishing to do this it is possible. I just saved the automation as my own then created a couple of helpers. One for the image url {{gif}} or {{image}} and one for the text response {{response.response_text}}. Just make sure your text response helper is maximum characters 255, and your LLM prompt is set to limit to 255 characters. Next, just below your notify action add:
- action: input_text.set_value
data:
value: "{{response.response_text}}"
target:
entity_id: input_text.frigate_event_label_front
- action: input_text.set_value
data:
value: "{{gif}}"
target:
entity_id: input_text.frigate_event_image_url_front
I then created hidden dashboards with these in a markdown card and directed the click action to them.
PS: Sorry for bastardising your beautiful blueprint 
Certainly leave it with me. Il check this out in the next few days.
Just a small update of the gemini models as this is what i use personally.
Iâm having trouble with this blueprint, the traces show that the analyzers error out with no files existing: "Error: Error: File downloader/my_frigate/1751393446.971977-mn39d9_snapshot.jpg does not exist
"
However, when I go to manually check for these files, they DO exist and I can access them.
Any suggestions???
Hi there so you can access the files from config/downloader/my_frigate/ or are you checking the frigate URL directly?
?
Just to confirm there is also no sub folders, after my_frigate.
On the trace does it continue after the error and then fail a second time after retrying the download.
Can you add a longer delay in say 30 seconds for the clip and maybe 10 seconds for the image to see if you get the automation to flow. If that works then dial back the delays until your happy with it.
Failing all that post your automation config or debug.
I am checking the files exist in the HA filesystem with terminal, and can access them at http://192.168.1.239:8123/api/frigate/notifications/1751393036.023949-uj3oia/snapshot.jpg where .1.239 is my HA machine (raspberry pi).
The downloader is putting the files in /config/downloader/my_frigate/ and there are no subfolder after that.
Here is a snapshot of the trace:
I tried to extend the maximum timeout, but that doesnât seem to be the issue. The automation seems to figure out when the file is done downloading, but it just cannot access it.
Please let me know what to try next, I really want this to work.
OKay looking at the trace, first thing I notice is the file has been downloaded and acknowledged.
They the downloader download action had no error and the event has been received to say the download is complete. The failure appears to be on the actual notification action.
First thing go check your JPG image is available in config/downloader/my_frigate folder and ensure the file is present.
Second thing to check if you can is copy the trace details,click the trigger when viewing the trace go to Changed variables and copy it all and paste here.
It exists, but the directory is /config/my_download_folder/my_frigate
Do I need to change âmy_download_folderâ to âdownloaderâ ?
Or can I change something in the Blueprint to point to my_download_folder?
Edit: Oh duh⌠Okay, I changed the âdownloaderâ name to my custom name and its working now 
THANKS!!!
Yes it should be what ever you set in your downloader integration.
Excellent.Thanks for the update.
Okay just looking over your requests
Under click actions you should already have options for the frigate clip directly or the downloaded image/clip.
Regarding your edit:
As the BP sits atm it can do this with the custom actions and just do what you need i.e.
action: input_text.set_value
metadata: {}
data:
value: "{{response.response.text}}"
target:
entity_id: input_text.bin_result
you could send the response to a media player tts for example.
the next update will include alot more options for the notification actions and optional action buttons etc.
Documentation Update
I really like your blueprint. I just need it to work with Signal notification service. I made this custom blueprint here to utilize the built in features of Frigate 16 BetaâŚhttps://community.home-assistant.io/t/sharing-my-blueprint-for-frigate-16-with-signal-alerts/902040
I would love it, if you could grab my blueprint and study it for merging it with yours. Your blueprint rocks!
Hi there, thanks for the upload I appreciate your interest.
What do you want to look at specfically?
I only assume its related to the frigate notification service, (which i have never used), not to discard this what is the reason for this as a backup method?
No, its related to the Signal service integration of Home Assistant. My blueprint would work with any notification service in theory.
