← Back to BrainyKit

A practical field guide

Use your sources.
Keep the receipts.

BrainyKit is a Windows desktop workspace for asking questions about your own files and web pages. This guide follows the app as it works today.

01 · Start here

Make one project for one body of work.

Create a project from the project switcher, then add a few related sources. A small, focused project is easier to ask about and easier to check than a mixed archive.

On first launch, open Models in the sidebar and download Medium or High. They are stored locally after download. Choose one and let it finish loading before sending your first question.

02 · Add your material

Bring in the documents you want to reason from.

Open Sources → Add source. You can upload PDF, Word (.docx), Excel (.xlsx), CSV, Markdown, or plain text files; paste text directly; or add a web link. BrainyKit extracts and indexes supported text so chat can find relevant passages.

For scanned PDFs, image-only pages, or documents with complex layouts, check the extracted text in the source preview. BrainyKit does not run OCR.

03 · Ask grounded questions

Ask a question that points to something in your project.

In Chat, ask about the project’s sources. Use the composer’s Strict mode when you want answers limited to what the sources say; choose Flexible when you want a plain-language explanation that can connect ideas while staying grounded.

Open a citation to inspect the passage it came from. If you want to focus a question on one source, use the composer’s + menu and choose Chat with a source. You can also attach a saved note or a file to a message.

To explore a reply from another angle, branch from a chat message. You can edit your earlier question, regenerate its answer, or stop a response that is still generating.

04 · Read and save useful passages

Keep the useful bits close to their source.

Open Study Board to browse project sources. Markdown files and saved web pages open in a reading view; select a passage to highlight it, ask about it, define it, or save it as a note. Other file types open in the source preview.

In Chat, save a whole answer or select part of a reply to save just that excerpt. Notes live in the project’s Notes section; saved follow-up branch trees appear under Branches. From a note linked to a chat message, use View in chat to return to its context.

05 · Make a study aid

Turn a source sample into a useful first draft.

Creation Studio can generate an overview, study guide, timeline, briefing document, mind map, diagram, flowchart, infographic, cheat sheet, or visual notes. You can scope a generation to selected sources and inspect the source excerpts used.

These outputs are starting points, not exhaustive summaries: the app samples representative chunks from the selected sources to fit the local model’s context limits. Check important details against the cited material.

06 · Choose a model

My usual pick is Gemma 4 E4B.

In my own use, I reach for Gemma 4 E4B most often, and recommend it if your computer has enough memory and you’re comfortable with slower responses. Gemma 4 E2B is my lighter, quicker fallback.

BrainyKit also has an optional Cloud mode for an OpenAI-compatible service. You provide its endpoint, API key, and model in Preferences; BrainyKit does not include cloud credentials. Switching to Cloud stops the local model process.

07 · Use voice tools

Speak a prompt, or listen back.

Use the microphone in the chat composer for dictation. Read-aloud controls are available on assistant replies, and Voices in the settings area lets you download and choose a supported voice. Voice downloads happen separately from the text model and downloading once unlocks all voices.

08 · Know what to double-check

Citations help you verify; they don’t guarantee correctness.

Open citations for claims that matter, especially names, dates, numbers, and decisions. A model can misread a passage or miss information that was not extracted. If the project does not contain the answer, add a relevant source or ask a narrower question.

Local mode keeps inference on your device after models are downloaded. Adding a web link requires an internet connection to capture the page; Cloud mode sends requests to the service you configure. Your files, notes, and projects remain part of the local app workspace.