- **NOTE:** Based on [initial investigation](https://gitlab.com/groups/gitlab-org/-/epics/12161#note_1676296356 "Code Suggestions: Connect IDEs through websocket to Instance"), we've opted to deprioritize the "Websocket to GitLab Instance" approach, in favor of connecting directly to the AI Model Gateway. - https://gitlab.com/gitlab-org/gitlab-vscode-extension/-/issues/1135+ We are moving forward with a bypass to the model gateway for the code completion use case to have improved latency there. ## ~~Keep a consistent websocket connection from IDE to Monolith~~ ~~To help us removing time for DNS + Instance Auth + etc. the idea would be to keep the constant connection to immediately send any code suggestions request through the final code completion logic+enrichment. As the logic for repository + user permission were already done before.~~ # ~~Steps~~ ## ~~1. Evaluate solution impact~~ ~~Lets evaluate the impact of the solution and POC a basic setup between Language Server and a GL instance over websockets that is kept consistent.~~