Per-model Combo Support #540
Replies: 3 comments 1 reply
-
|
Hey @vinz-bambico! Great idea - per-model combo routing is something I've been thinking about too Currently, combos are global (one active combo = one failover chain for all requests). But your use case - assigning different fallback chains per model/agent - is a strong one, especially for multi-agent setups like Oh-My-Opencode. Here's how you can achieve something similar today:
Thanks for the suggestion! |
Beta Was this translation helpful? Give feedback.
-
|
@dail45 -- here's the concrete how-to for the multi-key combo workaround today (before the per-model combo feature lands):
This is per-model-name routing, which is what OmniRoute supports today. The "true" per-model-combo feature (one global combo definition that re-dispatches based on the inner requested model) is tracked separately -- I'll link the new issue here when I open it. But for practical multi-agent setups, the alias-per-role approach above is what production users are doing and it works cleanly. |
Beta Was this translation helpful? Give feedback.
-
|
Update @vinz-bambico -- per-model combo routing is available now (it shipped via #563), so the manual one-combo-per-role workaround posted earlier is no longer required. Dashboard -> Settings -> Model Routing (
Backed by |
Beta Was this translation helpful? Give feedback.
Uh oh!
There was an error while loading. Please reload this page.
-
Given multiple agents with specialized skillsets, is it possible to make the combos be configurable on a per-model basis?
If I were to use Oh-My-Opencode for example, an ideal and more token-efficient scenario would be something like:
Sissyphus (main orchestrator agent): Claude Opus 4.6 -> Gemini 3.1 Pro -> Kimi K2.5 -> etc..
Hephaestus (code researcher): Claude Sonnet 4.5 -> GPT 5.3 Codex -> Gemini 3 Flash -> Kimi K2.5 -> etc.
Prometheus (planning agent): Claude Opus 4.6 -> Gemini 3.1 Pro GPT 5.3 Codex -> Kimi K2.5 -> etc..
You get the idea. Basically cascade providers to the next available as currently used one hit limits.
Since there is currently only one endpoint, I am thinking if it would be possible to create multiple endpoints for each specialized agent combo although I would leave it to your best judgement should you consider this idea.
Beta Was this translation helpful? Give feedback.
All reactions