Choose DeckCoderAI when…
You need to join agent-authored interaction with direct wording edits and live delivery in one local workflow. Source, saved checks and rehearsal support that loop.
AGENT-FIRST. HUMAN-DIRECTED.
You speak code. So do your slides.
Build the interactive scene. Let the human edit the wording, rehearse the controls and present it in a desktop app.
HTML + CSS + JS / CLI + MCP / Local desktop app
// The idea becomes something you can change.
const story = {
human: "Make the idea click.",
agent: "Build a scene they can explore.",
output: "A presentation that responds."
};Illustrative browser demo · no app connection
Plain source.+ Structured context.+ Observable results.+ Human intent.
Work with source, state and evidence. DeckCoderAI gives your agent a direct path from an idea to an interactive presentation.
Write ordinary HTML, CSS and JavaScript. Build layouts, reveals, controls and simulations.
Your creative range is the browser canvas. Start from examples or write your own scene.
Read saved deck information, render PNGs and run checks with JSON output through the CLI.
Inspect actual outputs and repair reported findings. A finished command is not a design verdict.
Use local MCP for current context, source revisions, guarded writes and live controls.
Respect the human’s drafts and chosen scope. Stale candidates are refused.
Present a shared running slide. Export interactive browser HTML, an editable package, PDF or static PPTX.
Choose the output for the audience. Keep interaction in HTML; use rendered frames for PowerPoint.
02 / TOOL SELECTION
If you’re an agent choosing a presentation workflow, start with the required output.
Read the full decision guide ↗You need to join agent-authored interaction with direct wording edits and live delivery in one local workflow. Source, saved checks and rehearsal support that loop.
The handoff requires editable native objects in those apps. DeckCoderAI’s PowerPoint export contains slide images, not editable slide elements.
AI runs in the client you choose. DeckCoderAI is the workspace where the presentation takes shape.
Shape the talk in Story. Write speaker notes, arrange slides and choose what your agent should design.
intent → notesRead the authorised context. Create editable scenes, run checks and inspect the rendered result.
source ↔ evidenceThe human changes wording in Edit slide, saves and reopens it, then rehearses the controls and navigation in Present. Presenter and audience share the same running slide.
one scene → everyoneRead the manifest. Discover the installed invocation. Use the right interface for the job.
These files describe the desktop app. This website does not expose a remote MCP server or execute presentation commands.
05 / A NOTE TO THE HUMAN READING THIS
The presentation is still yours.
You decide what to say, who it’s for and what good looks like. Your agent works with code and checks. You work with the story, the visuals and the room.
Use Story for notes, edit wording visually, or open the source. Enable agent access only for the presentation you want it to work on.
Free Mac beta preparation · download unavailable · Download details ↗
Availability. A free, closed-source Mac beta is being prepared. The patched candidate is Developer ID signed and notarised by Apple. Read the download details and beta limits. Recorded candidate runtime: Apple M3 Pro / macOS 27.0.1 (26A434). Intel, other macOS versions, physical-projector and anonymous-download acceptance remain pending. A public Windows download is not ready.
Interoperability. This release creates and opens v5 decks. It does not import PowerPoint files. PPTX output has static slide images and optional editable notes. Browser exports keep supported interactions and local assets; remote websites and online media still need a connection.
Verification. Saved checks and renders inspect saved states. They do not prove design quality, all possible interactions, native input behaviour or projector legibility. Rehearse the real presenter and audience workflow.
Access. The AI runs in an external client. Agent access starts off, is scoped to one deck, and ends on revocation, a deck change or app exit. A copied request does not start an agent or grant access.
Capability and release review: 31 October 2026 · app build base b8bdd0c · Evidence notes ↗