Model Fallback feature

[code-yeongyu]

Is there a way to have a model list that automatically switches to another model when the limit for OmO(Opus 4.5,) is reached?

For example, if Opus 4.5 reaches its limit,
OmO will automatically switch to using GPT 5.2.

Originally posted by @blacktoast in #37

I think general way of making fallback system through OpenCode Plugin SDK, and let OmO to use it would make sense.

The most tricky part is actually detecting the rate limit thing, because without sending the actual message there is no way to detect it.
In my personal experiences, using methods like invoking undo() from OpenCode and re-send if rate limited worked but worked like a house of cards.

But worth look into it.

[TomLucidor]

Picking smaller models as defaults would be great as well #585

[code-yeongyu]

Feature acknowledged. Model fallback system with automatic rate limit detection will be implemented as an enhancement to the OpenCode Plugin SDK. This will allow OmO to seamlessly switch between models (e.g., Opus 4.5 → GPT 5.2) when rate limits are reached. The tricky part of detecting rate limits proactively is noted - implementation will need to handle edge cases carefully.

[TomLucidor]

@code-yeongyu would love to see a more local-first/offline approach so people can pick custom services as well.

[unphased]

#1237! Let's go!

[code-yeongyu]

Enhancement Batch 22 — Classification

Category: Feature / Model Management
Priority: Medium
Effort: L — SDK-level feature with edge cases

Analysis

Model fallback on rate limit is a highly requested feature with multiple duplicate/related issues:

• Model Fallback #296 (Model Fallback — exact duplicate)
• Feature Request: Multi-model fallback per agent with automatic rate limit switching #326 (Multi-model fallback per agent with automatic rate limit switching)
• [Feature]: Support multiple fallback models for a single Agent (Model Failover Strategy) #703 (Support multiple fallback models for a single Agent)
• [Feature]: Configurable Model Resolution System #1026 (Configurable Model Resolution System)
• [Bug]: Model fallback system actually do not work!  #1540 (Models should also have a Priority Chain)

The core challenge remains: rate limit detection requires sending a message first, making proactive switching impossible. The undo() + re-send approach is fragile.

Recommendation

This issue should be consolidated with #296 as the canonical tracker. The implementation path depends on OpenCode Plugin SDK exposing error/retry hooks.

Duplicates

Closing in favor of #296 which is the canonical issue for this feature.

[code-yeongyu]

Closing as duplicate of #296 (Model Fallback). See also #326, #703, #1026, #1540 for related tracking.