Version and OS
0.55.8 on docker
Is your feature request related to a problem? Please describe.
Currently the settings only allow configuring a single LLM/AI provider globally. This forces every watch to use the same provider, which is a problem when you want to route different tasks to different models. I'm frustrated that I can't use a cheap local model for most watches while reserving a premium provider for the few that need it.
Describe the solution you'd like
Allow defining multiple provider configurations in settings (each with its own endpoint, model, API key, etc.), and add a per-watch dropdown to select which configured provider that watch should use. Optionally, allow marking one provider as the default for new watches.
Describe the use-case and give concrete real-world examples
I run small models locally, and they handle about 80% of my watches just fine. The remaining 20% need a stronger provider (Claude or ChatGPT) because the local models can't produce good results for them.
Example: dozens of simple watches (price/stock/text-change detection) run against a local Ollama endpoint. A handful of complex watches — where the page structure is messy or the extraction/summarization is harder — should be routed to the Anthropic or OpenAI API instead.
Being able to select the provider per watch lets me segment tasks by executor: local models for the bulk, paid APIs only where they're actually needed. This ultimately saves tokens and cost by not sending every request to a paid provider.
Additional context
Today the single-provider setting means it's all-or-nothing — either everything goes to the local model (and the hard 20% fail or degrade), or everything goes to the paid API (wasting tokens on the easy 80%). Per-watch provider selection solves both.
Version and OS
0.55.8 on docker
Is your feature request related to a problem? Please describe.
Currently the settings only allow configuring a single LLM/AI provider globally. This forces every watch to use the same provider, which is a problem when you want to route different tasks to different models. I'm frustrated that I can't use a cheap local model for most watches while reserving a premium provider for the few that need it.
Describe the solution you'd like
Allow defining multiple provider configurations in settings (each with its own endpoint, model, API key, etc.), and add a per-watch dropdown to select which configured provider that watch should use. Optionally, allow marking one provider as the default for new watches.
Describe the use-case and give concrete real-world examples
I run small models locally, and they handle about 80% of my watches just fine. The remaining 20% need a stronger provider (Claude or ChatGPT) because the local models can't produce good results for them.
Example: dozens of simple watches (price/stock/text-change detection) run against a local Ollama endpoint. A handful of complex watches — where the page structure is messy or the extraction/summarization is harder — should be routed to the Anthropic or OpenAI API instead.
Being able to select the provider per watch lets me segment tasks by executor: local models for the bulk, paid APIs only where they're actually needed. This ultimately saves tokens and cost by not sending every request to a paid provider.
Additional context
Today the single-provider setting means it's all-or-nothing — either everything goes to the local model (and the hard 20% fail or degrade), or everything goes to the paid API (wasting tokens on the easy 80%). Per-watch provider selection solves both.