Toolbox

​

What is Toolbox

• Runs in your browser. Code executes locally on your device. There is no server to set up and no external service processing your files.
• Private by default. Network access is disabled by default, so code running in Toolbox cannot send your data anywhere.
• Free with TypingMind. No extra subscription or API key is needed for Toolbox itself. You only pay for the AI model you already use.

​

Getting started

1. Click Plugins in the left sidebar.
2. Under Installed, select Toolbox from your plugin list.
3. Flip the toggle next to the Toolbox title so it turns blue.
4. Network access is disabled by default for this plugin. If you want to enable it, please switch to Settings tab to enable.

​

What Toolbox can do

​

How it works in a chat

1. You ask. Attach your files (Excel, CSV, PDF, Word, PPTX images, video, audio) and describe the outcome you want.
2. The AI plans and runs code. Each step appears in the chat as a tool call, for example Toolbox Run Python Inspecting PowerPoint template
3. It checks its own work. For bigger jobs you’ll see many small steps (“Check page space”, “Check watermark coordinates”, “Test copying pages”) before the final output. A “Worked for 2m 23s” label summarizes the run; click it to expand every step.
4. You get the result. Charts appear inline in the reply. Files appear as cards showing the file name, type, and size.
5. Previewing and downloading files: Click a file card to open it in the preview panel on the right. From there you can:

• Page through PDFs/slides and switch between sheets in a spreadsheet using the tabs at the bottom.
• Download the file with the download icon.
• Close the panel with the X.

​

Privacy, security and network access

• Where code runs: on your own device, in a sandbox inside the browser tab. TypingMind does not run your files on a server.
• What the AI model sees: the AI provider you chose still receives your messages and whatever the AI reads from your files in order to reason about them, the same as any chat. Toolbox doesn’t change your model provider’s data policy.
• Network access: disabled by default. With it off, code can’t download packages, call APIs, or fetch web pages. If you need data from the web, save the page or file and attach it instead (see the “Saved web page → pricing analysis” use case).
• Sensitive files: because processing happens locally, Toolbox is a good fit for financial data, customer lists, and internal documents. Still follow your organization’s rules about which AI models may see that data.

​

Practical use cases to try

​

Data analysis

Here's our sales export. Find the main trends and anomalies, then answer with visuals: build a 4-panel dashboard covering monthly volume, refund rate by month, average sales by day of week, and daily outliers. Explain each panel in 1–2 sentences2

We ran a new checkout A/B test in September. Is it a winner? Check statistical significance for conversion and revenue per visitor, break it down by device and traffic source, and give me a one-page PDF readout I can send to the product team.

This is a saved web page of a competitor's coffee catalog. Extract every product with price, origin, process, rating, and review count into a spreadsheet. Then chart price vs. rating and average price per 100g by process, and point out where we could price above or below them.

​

Documents, PDFs and reports

Combine these three documents into one board pack PDF. Add a beautiful cover page, next to cover, add a page with table of content, a short meeting-overview box, a footer on every page, and a light diagonal 'CONFIDENTIAL' watermark.

Use your_template.docx only as a style reference: fonts, heading colors, margins, and spacing. Using the attached Q3 financial summary PDF as the content, write a one-page Word memo to the board in that style, with a title, three key takeaways, and one small table. Don't reuse any of the contract text.

Turn these three board documents into a 5-slide PowerPoint for a 10-minute briefing: a title slide, one slide each for financials, roadmap, and hiring (one chart or table per slide), and a final slide listing the decisions we need from the board. Write each slide title as a takeaway, not a topic.

​

Images, Videos and audio

Place brewly_logo.png in the bottom-right corner of every photo at about 15% of the image width and 70% opacity. Then export each photo in three center-cropped sizes: 1080x1080, 1080x1920, and 1200x630. Put each size in its own folder inside one ZIP, and keep the original file names.

I've attached a 15-second vertical product video (1080x1920, no audio) and a music track. Make it a short ad for Reels:

1. Keep only the best part: trim the intro where the bottle sits alone and the final shot of the hand rubbing, so the result is 8 to 9 seconds centered on the dropper action. Tell me which time range you kept.
2. Add the text "Hydrating Serum" in the empty pink area at the top, in a clean white sans-serif font, fading in over the first 0.5 seconds. Add a second line below it: "Glow in 7 days".
3. Add the music as background audio. Detect the beats and put the cut or speed change on the nearest beat. Fade the music out over the last 1.5 seconds.
4. Export as MP4. Keep file under 10MB

turn these images into a 21-second loop GIF. Keep the file under 10MB. add TypingMind logo on the GIF on the top right corner. Under the logo - include a tag line - The best frontend for LLMs

Make a square audiogram video for LinkedIn from this podcast clip: animated waveform, the episode title "Ep. 42 · The 3 numbers every café owner should know", the show name Main Street Makers, and burned-in captions from the transcript.

Create a bar chart race video from this file showing the top 10 products by cumulative revenue, month by month from January 2025 to September 2026. 1920x1080, about 20 seconds long, with the month shown large in the corner. Export as MP4.

​

Tips for best results

• Name the deliverable. Say “give me an Excel file”, “a one-page PDF”, or “answer with visuals”. Without it, you may get a text answer only.
• Say who it’s for. “A readout I can send to the product team” or “for the board” changes the tone, length, and polish.
• Ask the real business question. “Did the January loyalty program improve retention?” gets a direct verdict, not just a chart.
• Explain your data. Mention what key columns mean, the date range, and known quirks (“refunds appear as negative amounts”, “the West file has a footer row”).
• Be precise with media. Give exact timestamps (“02:15 to 02:45”), pixel sizes or aspect ratios (“1080x1920”, “9:16”), the output format, and a target file size (“under 25 MB”).
• Ask it to flag problems. “Add a sheet flagging anything that looks off” surfaces duplicates, missing values, and mismatched totals.
• Iterate in small steps. Get the analysis right first, then ask for styling, extra sheets, or a PDF version.
• Pick a capable model for big jobs. Multi-step work (merging files, building PDFs, statistics) goes best with stronger models. Lighter models are fine for quick conversions.
• Check the numbers that matter. Expand the “Worked for…” steps to see what was calculated, and spot-check key totals against your source.

​

Limitations and troubleshooting

​

FAQs