Sliqtly better Slides
A lot of people seem to be asking how to convert, for example, Markdown files to PDF, PowerPoint or Word. The point being... it gets pretty expensive (and potentially slow) when the document goes through the same chat many times!
And .md files don't really have formatting. And what about charts, flowcharts, theming, not to mention pagination...
How Sliqtly compares
| Google Slides (API) | Claude + PowerPoint skill | Sliqtly | |
|---|---|---|---|
| Size of the deck | N/ALives in Google Drive | Fair91 KB .pptx for 15 slides | Excellent4 KB of Markdown for the same 15 slides |
| Works with any LLM | YesAny agent with Google API access and sign-in | NoClaude only | YesAny MCP client: Claude, ChatGPT, Cursor, VS Code… |
| Runs on your own machine | NoNo, Google's cloud | PartlyIn Claude Code yes; in claude.ai Anthropic's sandbox | YesYes: .deb, Homebrew, Docker |
| Changing it later | PoorNew API requests per change | FairRewrite the script, or edit slide and chart XML (12,000 tokens for 6 charts) | ExcellentAn edit of about 70 tokens; the deck reads back as 1,400 tokens of Markdown |
| Reusing the source data | FairMedium: charts can link to Sheets | PoorLow: the script is gone after the chat; chart data sits inside the .pptx | ExcellentHigh: Markdown, CSV and Excel kept with the deck, charts read them or a live Sheet |
| Speed | PoorAbout 100 API requests for 6 slides, each a round trip | GoodMinutes for a deck; about the same as Sliqtly | GoodMinutes for a deck; about the same, faster or slower from test to test |
| Look | FairGoogle's themes; every other style is set request by request | ExcellentThe most finished, native PowerPoint charts | GoodGood: themes, charts, diagrams, cards |
| Tokens the AI writes | PoorHigh: about 5,200 for 6 slides | GoodAbout 4,900 for 15 slides (the script); 7,600 of skill read first | GoodAbout 1,600 for 15 slides; but 22,000 of guide and tool definitions read first |
Measured 2026-10-07 on the same 15-slide deck (6 charts, 2 tables, a diagram, cards) made both ways; token counts with Anthropic's public tokenizer, which approximates current models. Google Slides is from our earlier 6-slide test; timings varied between tests, either way.
Example run
One deck made and then changed four times, in Claude, once with Sliqtly (cloud) and once with Claude's PowerPoint skill. Minutes per step; the first run in brackets.
| Step | Sliqtly (cloud) | Claude + PowerPoint skill |
|---|---|---|
| A, first version | 2.7 (1.8) | 3.6 (6.1) |
| E1, fix the data | 0.8 (1.2) | 0.3 (0.4) |
| E2, change the structure | 0.6 (1.0) | 0.7 (0.9) |
| E3, change the style | 0.5 (0.5) | 0.8 (1.1) |
| E4, export PDF | 0.15 (<1) | 0.1 (0.2) |
| Total | 4.75 (~5) | 5.4 (8.7) |
| Tool calls | ~42 | ~30 |
A deck and four rounds of changes, timed
We gave AI agents the same model and the same job on each platform: an 11-slide management deck from a customer spreadsheet (charts, a top-10 table, a flow diagram, a timeline, a risks table, speaker notes), styled after a reference picture. Then a manager's follow-ups: one wrong number fixed everywhere it shows, slides moved, deleted and added, the look changed, a PDF. Sliqtly makes the first version and changes the look fastest, and its edits change the deck in place. PowerPoint still looks the most finished, and its data fixes are quickest because a script regenerates every number. Sliqtly is behind on style control: fonts, card shadows and how a flow diagram is laid out.
| Claude + PowerPoint skill | Markdown → Pandoc → Beamer | Claude Slides | Sliqtly | |
|---|---|---|---|---|
| First version | Good3.6–6.3 min | Fair7.4 min, mostly raw LaTeX inside the Markdown | Poor19 min, most of it spent building a preview to check the slides | Excellent1.8–4.5 min in four of five rounds |
| Fixing one wrong number | Excellent0.3–0.5 min: one line in the script, every number regenerated | Fair1.5 min, 47 replacements | Poor9 min, bars and rings redrawn by hand | Good0.8–1.2 min, 18–27 find-and-replace edits in one call |
| Moving and adding slides | Good0.3–1.0 min, the whole file rebuilt | Good0.6 min | Fair3.9 min | Good0.3–1.1 min, in place; a move is a delete and an insert |
| Changing the whole look | Good0.8–1.1 min | N/ANot tested | N/ANot tested | Excellent0.5–0.6 min: a theme and a few lines of CSS |
| Match to the reference style | ExcellentRings, shadowed cards, framed pictures | GoodShapes right, text often small | ExcellentClosest, with the reference's font | FairColours and title right; grey cards, two fonts only, pale flow diagram |
| The agent checks its own slides | FairThrough LibreOffice | FairPDF to images | PoorNo tool for it | Excellentrender_overview and a layout report on every change |
| Shared as a link | NoA file | NoA PDF | YesPrivate until shared | YesA link; anyone with it can open the deck |
Five rounds on 2026-10-06 and 2026-10-07, one agent per platform on the same model, numbers checked against pandas on every platform. Times are minutes per phase. The last round with all four took 7.0 min in PowerPoint, 9.5 with Beamer, 11.3 in Sliqtly (server errors and a slide-deletion bug, now being fixed) and 33.5 in Claude Slides. Google Slides was left out: the connector lacked the Slides and Drive permissions. Marp and Slidev could not be installed in the test environment.
Features
What Sliqtly does today.
PowerPoint and Google Slides are a slightly unfair yardstick for a project this young: they have decades of work behind them. Here is plainly what Sliqtly reads, draws and does today.
Charts
Vega-Lite, drawn by Sliqtly's own Vega engine (46 chart types tested against Vega):
- bar, stacked, grouped and normalized bar, histogram
- line, multi-series line, area, streamgraph
- scatter, bubble, heat map, tick
- pie, donut, text labels, layered, faceted and side-by-side charts
- time and log scales; filter, calculate, aggregate, bin, window, fold, pivot, lookup, regression, loess, density
- looks: flat, forge, neon, glass, cartoon, romantic
- chart text in the deck's language (lang:)
Not yet: errorbar, errorband, impute, sample.
Diagrams
- Mermaid: flowchart, sequence, class, state, ER, mindmap, timeline, journey, gantt, gitGraph, pie, quadrant, xychart, sankey, block, architecture, kanban, requirement, C4 context, packet, radar, treemap
- PlantUML: sequence, class, object, component, deployment, use case, activity
- Graphviz DOT, D2
- a guided tour, box by box
PlantUML state, timing, mind map, Gantt are not drawn.
Diagram looks
- classic
- sketch (hand-drawn)
- mermaid (pastel cards)
- jurassic (poster)
- cartoon (speech bubbles)
- romantic (caption boxes)
Layouts from lists
- process (chevron steps)
- SWOT
- timeline
- cards
- stats (key figures)
- gallery (photo albums)
Text
- tables, with rowspan/colspan and toned cells
- math in LaTeX, inline and display
- code with highlighting, line numbers and stepped lines
- diffs
- build steps, transitions, speaker notes
Pictures and art
- SVG
- PNG, JPEG, GIF, WebP
- SmartArt (partial: 20 layouts)
- a picture as a slide background
Data
- inline values
- CSV and JSON files or URLs
- Excel workbooks kept with the deck
- Google Sheets, read live
- your own keys in the front matter, shown in headers and footers
Output
- the player (HTML, a link)
- PDF, also for print with bleed and crop marks
- PowerPoint (.pptx, editable text, formulas as equations)
- Word (.docx)
- a web page (.html)
- Markdown, with or without review comments
Themes
- aurora, nebula, carbon, ember, midnight (dark)
- white, corporate, pearl, hive, lattice, apex, tide, mist (light)
- forge, foundry, site (work)
- clinic, care, vital (health)
- editorial (A4 pages)
- your own CSS on top
Books and albums
mode: book: pages in spreads, inside and outside margins- a realistic 3D page turn in the viewer
- headers, footers and page numbers
AI assistants (MCP)
- any MCP client: Claude, ChatGPT, Cursor, VS Code…
- create, edit and export decks; read back a layout report
- render slides to check them
- bind charts to data, write Excel workbooks
- list, add and resolve review comments
- sees what you changed by hand since its last look
Editor
- at sliqtly.com/editor after Google sign-in
- Markdown and the slides side by side
- diagram and table windows for editing blocks
- version history
- review mode: comment pins and threads on slides
- draw on slides, record voice and pointer while presenting
- dark and light looks
Working together
On your own Sliqtly server:
- rooms (projects) holding decks, folders and files
- room chat with threads, files and link previews
- editing the same deck together, live
- voice calls inside a room
Sliqtly Personal
Wanna run local?
The Sliqtly server for Ubuntu and Debian, macOS (Homebrew) and Docker: the MCP endpoint for your agents and a viewer for your decks, on your own computer. Data stays between your machine and your agent.
sudo apt install ./sliqtly-personal_amd64.deb
Free for personal use and qualifying solo businesses with annual gross revenue below €200,000. All other commercial use requires a commercial license. License
Who?
Just a software engineer working at a company called Cirit.
I build software for a living, but this is my passion.