I created an open source Copilot web extension for Azure Devops PRs

[mshenawy22]

Select Topic Area

Question

Body

Hey Everyone,

I have created a copilot chrome extension 'Thinkreview' that integrates within ADO PR pages , generated comprehensive code reviews , allows you to chat with the PR to understand the code changes and generate comments to post on the PR .
All of this without leaving the Azure PR page

The project is open source on github and I'd like anyone suffering from the same pain point to give it a try
you can also configure which llm model you use and it supports ollama

https://github.com/Thinkode/thinkreview-browser-extension

https://chromewebstore.google.com/detail/thinkreview-ai-code-revie/bpgkhgbchmlmpjjpmlaiejhnnbkdjdjn

[krishkhinchi]

It's a good idea to incorporate AI-assisted reviews straight into the Azure DevOps PR page to solve a significant workflow obstacle. For larger or unfamiliar codebases, the ability to communicate with the PR and produce review comments in context is particularly helpful. For teams with budgetary or privacy concerns, supporting Ollama and configurable LLMs is a wise design decision. I'm interested in how you handle prompt limits and different context sizes for large PRs, as well as whether reviewers can alter the review criteria. All things considered, this appears to be a useful addition to the ADO ecosystem.

[mshenawy22]

To answer your question regarding prompt limits and large PRs, our system manages context through a three-layered approach:

Model-Specific Context Windows: We support a variety of LLMs with different capacities. You can choose a model that fits your PR size requirements; a full list of supported models and their context limits can be found here.

Smart Truncation Logic: Our backend automatically filters out 'noise' to prioritize meaningful code. The system excludes less relevant files and assets—such as lock files, media, node_modules, , minified code and others —ensuring the LLM focuses only on the logic that requires review.

LLM Caching: We utilize advanced caching techniques to maximize the available context window. This not only allows us to process larger diffs but also significantly reduces latency and API costs.