Engel Nyst

Quick fixes for Agent readme.

The typewriter feature is disconcerting when responses are longer than a few lines

3

**Describe the bug** The chat with OpenDevin the AI now displays its thoughts character-by-character, which creates a cool impression of thinking as it "types". But that impression breaks when the...

Default to a less expensive model

6

In relation to https://github.com/OpenDevin/OpenDevin/issues/449 This PR proposes a simple measure for unexpected costs: default to GPT-3.5. Everyone who wishes GPT-4 can choose it in configuration. Note that it changes also...

The default model from the UI is not applied

4

**Describe the bug** The UI has now the ability to choose model, and it defaults to GPT-3.5. But GPT-3.5 is not the model in use, *if you had another in...

Set up pytest with pipenv

Remove md around json

2

We had multiple reports of broken JSON from the LLM. I think the presence of markdown in prompts must be what is "breaking" responses in this particular way: the LLM...

Add logging

1

This is WIP rebased on top of the switch to poetry branch. Sorry it's harder to see diff this way; that will be fixed. @dorbanianas introduced logging on the testing...

Small refactoring as follow-up to json cleaning: - move another function that was guarding responses from "bad" json. Once moved, it seems to me that it is redundant? I think...

Upon more investigation on this one, it turns out that: - our most active upstream, liteLLM, released a patch and fixed some issues with async calls and their Router class.([example](https://github.com/BerriAI/litellm/commit/0d1cca9aa0a853452fa67032e98e670fc43c8598))...