LOCAL AI / RAG STEP BY STEP

Local RAG Lab

Build a working RAG pipeline entirely in your browser. Run each stage separately and inspect the code, operation log, and output passed to the next stage.

Start the labView the workflow
8 stagesfrom environment setup to source checking
1 local AI engineinitialized before documents are loaded
2 execution panelscode and live operation log
no hidden transitionsNext unlocks only after success

GUIDED WORKFLOW

From browser setup to an answer with source citations

The panel on the right shows the workflow and the status of every stage. The workspace displays completed stages and one active stage; future stages open when you select Next.

1 / 8

Step 1. Choose a model and configure the execution environment

First choose where the generative model will run. WebGPU mode uses WebLLM and GPU acceleration and offers several compact models. When WebGPU is unavailable, the lesson switches to Transformers.js and WebAssembly. CPU mode offers two models: multilingual Qwen2.5 0.5B and the smaller, English-oriented SmolLM2 360M. CPU generation is considerably slower, but it lets the complete RAG workflow run without a compatible GPU. The lab then checks the browser, reports every library and model download, starts a Worker, and performs a real generation smoke test.

Parameters

Current stage
Automatic selectionWith WebGPU, the lab selects a compact WebLLM model. Without WebGPU, it uses CPU / WebAssembly: SmolLM2 360M for English and Qwen2.5 for Russian or Chinese.

Code being executed

    Operation log

    This stage has not run yet.
    1. This stage has not run yet.

    Stage result

    This stage has not run yet.

    This stage has not run yet.