GOAL Build a private search engine for the documents folder in this project. I ask a question in my own words and get the passages that answer it, even when my words differ from the words in the file. Then I can ask for a short answer written only from those passages, with a source for every claim. Everything runs locally with Ollama. DONE MEANS For my five test questions, the correct source passage appears in the top five results. Answers cite file name and page or section for every claim. When the answer is not in my files, the app says so instead of guessing. Adding a new file updates the index without rebuilding everything. HOW TO WORK Work through every stage below without waiting for me. Verify each stage with a real test before you start the next. Stop only when you need something only I can do: give test questions, judge a result, or approve a change. Explain each handoff in plain English. Keep a TEST_LOG.md with every check, its result and the exact error if it failed. STAGE 1: INSPECT AND PREPARE Inspect my operating system, CPU, RAM, GPU and free disk space. Record them. Check that Ollama is installed and running; if not, guide me through the official installer. Create an isolated Python environment for this project only. Pull EmbeddingGemma (embeddinggemma) for embeddings. Read its current model card and record the exact input format it expects for search queries and for documents; use those formats. Pick a local chat model from the Ollama library that fits my RAM and record why. Install Chroma as the vector database, stored on disk in this project. Record all versions. STAGE 2: EXTRACT AND SPLIT Extract text from every file in the documents folder (TXT, MD, PDF, DOCX). Keep the file name, page number or heading, and file modified date with every piece of text. List files that produced no text, such as scanned PDFs, and skip them with a clear note. Split the text into passages of a few hundred words, at paragraph boundaries, with a small overlap so an idea that spans two passages is not lost. Save a sample of passages to a file I can read. STAGE 3: EMBED AND STORE Turn each passage into an embedding (a list of numbers that represents its meaning) with EmbeddingGemma. Store each embedding in Chroma with its text, file name, page and a hash of the file. Use the same embedding model for documents and questions, always. Make indexing incremental: on each run, embed only new or changed files and remove passages from deleted files. Show the number of files, passages and the time taken. STAGE 4: MAKE SEARCH GOOD FIRST Build search: embed my question with the query format, retrieve the top results from Chroma, and show each passage with its file, page and similarity score. Add an Open file button for each result. HANDOFF: ask me for my five test questions and the file each answer is in. Run them. Report for each question whether the right passage was in the top five. If fewer than four pass, adjust passage size and overlap, re-index, and rerun. Record each attempt and its score in the test log. Do not add the chat model until search passes. STAGE 5: ADD GROUNDED ANSWERS Add an Answer button. Send the chat model my question and only the top retrieved passages, each labelled with its source. Instruct it to answer only from those passages, to cite the source label after every claim, and to reply "I could not find this in your documents" when the passages do not contain the answer. Show the answer above the passages it used, so I can check it. Limit the passages sent to what fits comfortably in the chat model's context window. Show how many passages were used. STAGE 6: MAKE IT EVERYDAY Create a simple local web page with a search box, results, the Answer button and a Re-index button. Show when the index was last updated. Let me change the documents folder in settings. Keep the database in a visible folder outside the build files. Run every acceptance test in this pack. Mark each PASS, FAIL or NOT RUN in TEST_LOG.md. Repair failures before you add any feature. DELIVER The command or launcher I use to start the app, the project folder, the database folder, the models and versions used, my test questions with their scores, a requirements lock file, a short README and the test log. List which checks still need my judgement. Keep everything local; no cloud APIs and no telemetry.