I have two Ollama based Voice Assistants. One Gaming PC and another server with a smaller GPU.
I couldn’t get the pipeline to fall back to the secondary Ollama VA when it was offline. Satellites just went dead. I see a few other people had this issue too. I thought about using a load balancer / proxy at the network/http layer and maybe I’ll get there.
But for now I have an automation which periodically does a health check of the Preferred Ollama server, and if unavailable, sets the ESP Satellite index to select the other VA. It reverses this when the preferred one returns.
Now, when I shutdown my PC the kids can still “Hey Mycroft” the TV on in the morning 
Though it might be useful. Feel free to offer suggestions. It looks a little complicated for multiple Satellites and edge cases. But it’s straightforward in architecture and implementation.
It assumes you have ESP Home Satellites. I haven’t tried on Wyoming. Only other requirement is a shell script in your configuration.yaml to curl the Primary Ollama Server status.
Need to fix:
- Doesn’t work for Companion App, where user appears to have to select VA manually.
- I can’t for the life of me figure out how to rename my states(‘select.my_assistant_assistant_index’) id - the first name I registered my Satellite with (it’s now, “Pi4” in every other place I can think of including .storage but it remains cached).
Shell command
shell_command:
# Poll Ollama preferred server for health / availability
ollama_preferred_hc: "bash -c 'curl -s --max-time 5 http://192.168.0.XXX:11434/ || echo Server is down; exit 0'"
Automation
alias: Ollama Health Check
description: >
Periodic health check for Ollama server and device selection based on
response.
triggers:
- minutes: /5
trigger: time_pattern
conditions: []
actions:
- data: {}
response_variable: ollama_preferred_hc
action: shell_command.ollama_preferred_hc
alias: Poll Preferred Ollama Server WIth curl
- data_template:
name: Debug Ollama Health Check
message: >
Ollama: {{ ollama_preferred_hc.stdout }}; Select1: {{
states('select.pi3_assistant_index') }}; Select2: {{
states('select.my_assistant_assistant_index') }}
action: logbook.log
alias: Log Ollama and Satellite States
- alias: Fallback to backup Ollama VA if Needed
choose:
- conditions:
- alias: Preferred Ollam is Down
condition: template
value_template: |
{{ ollama_preferred_hc.stdout | trim == "Server is down" }}
sequence:
- device_id: 5446524fd34986e8a2abbb953d210515
domain: select
entity_id: 7c81903a1fb7c6e5b1b018e3971c03f5
type: select_option
option: Ollama 3070ti
alias: Set Pi4 ESP Satellite VA to backup Ollama 3070ti
- device_id: 59caf315ff7e210bbf03fd5aba73051b
domain: select
entity_id: 6ee96beaf3f91624d88c67553ffc538f
type: select_option
option: Ollama 3070ti
alias: Set Pi3 ESP Satellite VA to backup Ollama 3070ti
alias: Preferred Ollama is down - Fallback to backup
- conditions:
- condition: and
conditions:
- condition: template
value_template: |
{{ ollama_preferred_hc.stdout | trim == "Ollama is running" }}
alias: Preferred Ollama is running
- alias: >-
Any Satellite State != "preferred" (Ignoring unavailable and
unknown)
condition: template
value_template: >
{% set s1 = states('select.pi3_assistant_index') %}
{% set s2 = states('select.my_assistant_assistant_index') %}
{{ ((s1 != 'preferred') and (s1 != 'unknown') and (s1 !=
'unavailable')) or
((s2 != 'preferred') and (s2 != 'unknown') and (s2 != 'unavailable')) }}
sequence:
- device_id: 59caf315ff7e210bbf03fd5aba73051b
domain: select
entity_id: 6ee96beaf3f91624d88c67553ffc538f
type: select_option
option: preferred
alias: Set Pi3 ESP Satellite VA to Preferred
- device_id: 5446524fd34986e8a2abbb953d210515
domain: select
entity_id: 7c81903a1fb7c6e5b1b018e3971c03f5
type: select_option
option: preferred
alias: Set Pi4 ESP Satellite VA to Preferred
alias: >-
Preferred Ollama is Up & Satellites are Still Pointing at Backup -
Restore
mode: single
Picture if the yaml is a bit rough on the eyes…
