context related suggestions

[asukaminato0721]

Validations

• [X] I believe this is a way to improve. I'll try to join the Continue Discord for questions
• [X] I'm not able to find an open issue that requests the same enhancement

Problem

take markdown as an example

suppose I have

[xx](https://example.com/xx)

then I edited it

[xx1](https://example.com/xx)

I hope it can suggest changes like

[xx1](https://example.com/xx1)
                            ^ here

like F2 but more intelligent.

Solution

code completion like this.

[Patrick-Erichsen]

That would be useful! I'm not certain if the issue is with model capabilities or with the context we are providing. Are there any other extensions you've used that have given better suggestions in a scenario like that?

[asukaminato0721]

Are there any other extensions

The solution is from cursor. ;-)

---

and another relate example:

suppose

1,2,3

- 1
- 2
- 3

auto complete it to

1,2,3,4

- 1
- 2
- 3
- 4 <- here

[Patrick-Erichsen]

This is more in line with "next intent" prediction than the current Autocomplete we've built. Definitely on our radar in the near future! Appreciate the feedback here though.

[TechxGenus]

Copy from discussion: https://discord.com/channels/1108621136150929458/1108621136830398496/threads/1294611370045603933

---

I previously worked on attempting to implement core features of Cursor (see https://github.com/TechxGenus/CursorCore).

Using existing base models or instruction models may not be suitable, as they are not aligned with this 'automated editing' task. They lack corresponding training so thier performance may be suboptimal. When trying to achieve similar results using prompts, a large amount of information must be included - leading to lengthy inputs and significant latency.

Cursor claims it is powered by a fine-tuned custom model, and I believe fine-tuning is currently the best approach. I've experimented with training some models. As for various diff formats, there's a trade-off between performance and speed, and I think the choice should depend on specific contexts.

I've developed a simple web-based simulation of similar functionality, and now I'm considering how to integrate it into code editor tools like continue (unfortunately I'm not familiar with ts and continue sourse code). Maybe we could open a new issue to discuss this further?

[Rogeriohsjr]

+1 on this feature