The main components of Developer Lightspeed are the large language model (LLM), a Visual Studio Code (VS Code) extension, and the solution server.

When you initiate an analysis, Developer Lightspeed creates a context to generate a hint or a prompt that is shared with your LLM. The context is drawn from profile configuration that contains the target technology or the source technology that you configure for migration. Based on the source or target configuration, Developer Lightspeed checks the associated rule set containing rules that describe what needs to be fixed as the first input for the LLM prompt. As the second input, Developer Lightspeed uses the solved examples that are stored in the solution server from previous analyses in which the source code changed after applying the fixes.

The Solution Server uses Model Context Protocol (MCP) to act as an institutional memory that stores changes to source codes from analyzing all the applications in your organization. This helps you to leverage the recurring patterns of solutions for issues that are repeated in many applications. Solution Server reuses these past solutions through fix suggestions in later migrations, leading to faster, more reliable code changes as you migrate applications in different migration waves. At the organizational level, the solution server can also help Developer Lightspeed tackle a new issue given how a similar issue was resolved in a previous analysis.

The hint or the prompt generated by Developer Lightspeed is the well-defined context for identifying issues that allows the LLM to "reason" and generate the fix suggestions. This mechanism helps to overcome the limited context size in LLMs that prevents them from analyzing the entire source code of an application. You can review the suggested change and accept or reject the update to the code per issue or for all the issues.

Developer Lightspeed supports different goals of analysis through the three modes: the Agentic AI, the Retrieval Augmented Geeneration (RAG) solution delivered by the solution server, and the demo mode.

If you enable the agentic AI mode, Developer Lightspeed streams an automated analysis of the code in a loop until all issues are resolved and changes the code with the updates. In the initial run, the AI agent:

If you accept that the agentic AI must continue to make changes, it compiles the code and runs a partial analysis. In this phase, the agentic AI can detect diagnostic issues (if any) generated by tools that you installed in the VS Code IDE. You can accept the agentic AI’s suggestion to address these diagnostic issues too. After every phase of applying changes to the code, the agentic AI runs another round of automated analysis depending on your acceptance, until it has run through all the files in your project and resolved the issues in the code. Agentic AI generates a new file in each round when it applies the suggestions in the code. The time taken by the agentic AI to complete several rounds of analysis depends on the size of the application, the number of issues, and the complexity of the code.

The RAG solution, delivered by the Solution Server, is based on solved examples or past analysis to fresolve new issues or similar issues that are found while analyzing the source code. This type of analysis is not iterative. The Solution Server analysis generates a diff of the updated portions of the code and the original source code for a manual review. In such an analysis, the user has more control over the changes that must be applied to the code.

You can consider using the demo mode for running Developer Lightspeed when you need to perform analysis but have a limited network connection for Developer Lightspeed to sync with the LLM. The demo mode stores the input data as a hash and past LLM calls in a cache. The cache is stored in a chosen location in the your file system for later use. The hash of the inputs is used to determine which LLM call must be used in the demo mode. After you enable the demo mode and configure the path to your cached LLM calls in the Developer Lightspeed settings, you can rerun an analysis for the same set of files using the responses to a previous LLM call.