Support separate models & mechanism for applying changes

[anh-chu]

What specific problem does this solve?

• Problem: apply_diff intermittently fails with various models
• All users are affected
• This happens when models are trying to edit files with apply_diff, but particularly more with setups like automatic linting/prettier. Seems to also happens more with multiple continuous edits where (presumably) the models become more flaky with context. I am using both Claude Sonnet 4 and Gemini Pro 2.5 Preview.
• Current behavior: the chat/edit models use apply_diff themselves, requiring strict diff format and adherence to existing file contents while also diluting the context window.
• Expected behavior: a separate apply model such as Morph Fast Apply or Relace’s Instant Apply model can be chosen to apply changes instead, reducing the burden of generating lengthy and correct diffs.
• Impact: the current methods consume more tokens, enlarge the context, and can be more unreliable and time-consuming. With a new apply method, we can expect stable, long running flows with fewer re-edits.

Additional context (optional)

This is inspired by continue.dev's paradigm, which literally never failed me while making complex edits. Some links:

• https://blog.continue.dev/instant-apply/
• https://docs.continue.dev/customize/model-roles/apply

Roo Code Task Links (Optional)

No response

Request checklist

• I've searched existing Issues and Discussions for duplicates
• This describes a specific problem with clear impact and context

Interested in implementing this?

• Yes, I'd like to help implement this feature

Implementation requirements

• I understand this needs approval before implementation begins

How should this be solved? (REQUIRED if contributing, optional otherwise)

No response

How will we know it works? (Acceptance Criteria - REQUIRED if contributing, optional otherwise)

No response

Technical considerations (REQUIRED if contributing, optional otherwise)

No response

Trade-offs and risks (REQUIRED if contributing, optional otherwise)

No response

[roomote]

I am working on implementing support for separate models and mechanisms for applying changes. This will include adding configuration options for dedicated apply models like Morph Fast Apply or Relace's Instant Apply model to improve reliability and reduce token consumption when making file edits.

[anh-chu]

Attached is an example of how a task with unreliable edits looks like. It was straightforward until editing one specific files that looped for around 10 times before I manually intervened. The token use and task time was significantly inflated. When I tested with Continue.dev using the task planning context info from Roo, it was done without any similar issues.

Model used: Gemini 2.5 Pro Preview 06-05, standard Code mode.

[roomote]

I see that @roomote is already working on this issue. I am also assigned to work on implementing support for separate models and mechanisms for applying changes. I will coordinate to ensure we are not duplicating efforts and will focus on complementary aspects of the implementation if needed.

[daniel-lxs]

Thanks for the thoughtful proposal and the detailed context - it's clear you've put a lot of thought into this.

At the moment, we don't have plans to introduce a separate apply model flow like the one you're describing. While we understand the challenges with diff-based editing, especially across multiple edits or with strict formatting setups, our current focus is on improving the reliability of the existing apply_diff pipeline rather than introducing a parallel apply model architecture.

That said, we’re always open to feedback and evolving based on user needs. We’ll keep an eye on this and revisit if it becomes a more common request. Thanks again for raising it.