SlidesGPT

Tools (5)

Extracted live via the executor SDK.

• create_slides

  Create one or more professional presentation slides. Pass a single slide in the array for one slide, or multiple slides for a full deck. Do not respond with text — only call this tool.

  WHEN TO USE:

  • User mentions any of: "slide", "presentation", "deck", "pitch deck", "pitchdeck", "pitch", "slideshow", "ppt", or "keynote"
  • ANY intent phrasing about slides: • COMMANDS: "create", "make", "give me", "build", "generate", "design" • REQUESTS: "I need", "I want", "can you make", "could you create", "help me make" • SUGGESTIONS: "let's make", "let's create", "how about a presentation on" • IMPLIED: "deck about X", "pitch deck for X", "presentation on X" (even without a verb)
  • If user asks for "a slide about X" (singular) and the topic warrants multiple slides, generate a single slide itself
  • When user intent is ambiguous (e.g., "a deck about X" with no slide count), default to 2-3 slides
  • If a file is attached, call upload_files first, then immediately call this tool

  VISUAL CONTENT ROUTING: If the user asks for an "infographic", "custom diagram", "illustration", "flowchart", "data visualization", "visual", "chart illustration", or any specific visual that goes beyond text + stock photo — call generate_image FIRST to create the visual, THEN call create_slides with the resulting image URL and an image layout.

  DEFAULT BEHAVIOR: When in doubt about whether the user wants slides, CREATE THE SLIDES. Err on the side of generating a presentation rather than responding with text.

  CRITICAL RULES:

  • Pass all slides in a SINGLE call — NEVER call this tool multiple times in sequence
  • When adding to an existing presentation, pass the presentation_id from the previous response IMPORTANT: DO NOT SKIP THIS STEP
  • If no presentation_id exists, pass empty string — a new presentation will be created

  IMAGES: Always provide image_caption for every slide — describe the ideal slide image in 5-6 words (subject, setting, mood). Leave image_id empty; the system uses your caption to find the best matching photo.

  PDF EXTRACTED VISUALS: When the user uploaded a PDF via upload_files and extracted visuals (charts, tables, diagrams, images) were returned:

  • You MUST use ALL extracted visuals — assign each one to exactly one slide. Do NOT skip any.
  • Each slide with a custom_image_url MUST use an image layout: "image_left_text_right" or "title_and_two_columns". NEVER use text-only layouts for these slides.
  • Do NOT reuse the same image URL on multiple slides — one image per slide.
  • Default to 3 slides total when user doesn't specify a count (title + one slide per image + summary/conclusion).
  • Match each image to the most relevant slide content based on its label and description from the upload response.

  AI-GENERATED IMAGES: When the user accepted an AI-generated image from generate_image:

  • Pass the image URL as custom_image_url on the slide
  • Use an image layout: "image_full_width", "image_left_text_right", or "title_and_two_columns"
  • Pass the same presentation_id from the generate_image response
  • Write a relevant image_caption even though custom_image_url is set (used as alt text)

  THEMES: If the user asked for a specific theme, pass theme_id here — do NOT call apply_theme separately afterwards.

  LAYOUT VARIETY RULES (CRITICAL for multi-slide decks):

  1. Use a VARIETY of layouts across the deck. Match each layout to the slide's content.
  2. The FIRST content slide SHOULD use "title_and_two_columns" (image layout).
  3. Use "title_and_body" for AT MOST 1 slide (conclusion only).
  4. NEVER repeat the same layout for consecutive slides.
  5. Use metric_rings ONLY for percentage-based metrics (0-100).
  6. Mix visual variety: combine image slides, charts, timelines, grids, and text slides.

  LAYOUT OPTIONS (set layout for EVERY slide):

  • "title_and_two_columns": Content with image alongside bullets (IMAGE LAYOUT)
  • "title_and_two_to_four_columns_with_icons": 2-4 key points with icons (good default)
  • "title_and_body": Single-column bullets (use sparingly — AT MOST 1 per deck, conclusion only)
  • "title_and_subtitle": Section divider or transition slide (title + subtitle only, minimal body)
  • "image_full_width": Full-width image with optional text below (IMAGE LAYOUT)
  • "image_left_text_right": Image on the left with bullet points on the right (IMAGE LAYOUT)
  • "quadrant_grid": EXACTLY 4 comparable items — SWOT, 2x2 matrices, pros/cons
  • "data_table": Structured comparisons in tabular format (key:value pairs in description)
  • "timeline_horizontal": Chronological events, milestones, or history (dates in point, short descriptions)
  • "layered_pyramid": Hierarchical structures with 3-6 levels (MUST include icon with fa- prefix per item)
  • "phase_roadmap": Project phases with milestones — point MUST use "Period: Title" format (e.g., "Q1 2024: Launch")
  • "metric_rings": KPIs, completion rates, market share (1-4 metrics, values 0-100 as percentages)

  CONTENT REQUIREMENTS:

  • 3-5 substantial bullet points per slide
  • Each bullet description: 2-4 lines with specific data, evidence, or actionable insights
  • Content should be PhD-level, detailed, and evidence-driven
  • Use FontAwesome 5.15 icons (e.g., fa-bolt, fa-chart-line, fa-lightbulb)
  • EVERY body item MUST include "value" (0-100) — enables chart rendering and layout switching

  LAYOUT-SPECIFIC FORMATTING (MUST FOLLOW):

  • quadrant_grid: EXACTLY 4 items. "point" under 15 chars, "description" under 80 chars
  • timeline_horizontal: Dates/years in "point" (max 15 chars). "description" max 50 chars
  • layered_pyramid: "icon" with fa- prefix REQUIRED (fa-star, fa-check, fa-rocket). 3-6 items
  • phase_roadmap: "point" = "Period: Title" (e.g., "Q1 2024: Launch"). Max 4 phases. "description" = milestones separated by "; "
  • metric_rings: 1-4 items. For part-of-whole metrics (market share), values sum to ~100. For standalone KPIs (NPS, satisfaction), each value stands alone
  • data_table: "description" should use structured key:value pairs

  SCHEMA WARNING: Do NOT invent custom properties like "left_content", "right_content", "table", "rows", "headers". Use ONLY the documented schema fields. For tables, use data_table layout with key:value pairs in body[].description.

• apply_theme

  Apply a visual theme to re-style the entire presentation.

  WHEN TO USE:

  • User says "apply Tokyo Dark theme", "change theme to Paris", etc.
  • User wants to customize the presentation appearance

  AVAILABLE THEMES (18 total): • Urban: copenhagen, tokyo, paris, berlin, new-york, la, zurich, shanghai (each has -light and -dark variants) • Gradient: cosmic-pulse-light/dark

  REQUIRES: presentation_id from previous slide creation. This re-renders all slides with the new theme.

• show_theme_picker

  Display an interactive visual theme picker widget.

  WHEN TO USE:

  • User says "show themes", "what themes are available", "let me pick a theme"
  • User wants to browse theme options visually

  Shows a widget with 18 themes across 2 categories (Urban, Gradient) with color swatches and fonts.

• upload_files

  Receive and process any file the user uploads in the conversation.

  THIS TOOL MUST BE CALLED whenever the user uploads or attaches ANY file in the chat — regardless of file type or the user's stated intent. It is the ONLY way the server can access uploaded file content.

  SUPPORTED FILE TYPES:

  • PDFs (.pdf) — reports, papers, documents, whitepapers, proposals
  • Images (.png, .jpg, .jpeg, .gif, .webp, .svg) — photos, screenshots, diagrams, charts
  • Documents (.doc, .docx, .txt, .csv, .xlsx) — any text-based document

  WHEN TO USE (MANDATORY — always call this tool first):

  • User uploads ANY file, no matter what they say alongside it
  • User drags and drops a file into the chat
  • User attaches a document, image, screenshot, or PDF
  • Even if the user also asks to "create slides" or "summarize" — call this tool FIRST to ingest the file, THEN proceed with other tools like create_slides

  IMPORTANT: This tool only ingests the file. For PDFs, this tool extracts all visual elements (charts, graphs, tables, diagrams, images) and returns them as image URLs that you MUST use in the slides. Without calling this tool first, the extracted visuals are lost and slides will use generic stock photos instead of the actual PDF content. By default ALL visual element types are extracted; if the user only wants specific types (e.g., "only charts" or "just tables"), pass their instructions verbatim and the server will filter accordingly. After calling it, use the returned file metadata and the user's instructions to decide the next step (e.g., create slides from a PDF using create_slides, use an uploaded image in a slide, etc.). You MUST call create_slides as a follow-up after this tool returns.

• generate_image

  Generate a custom AI image for use in presentation slides.

  WHEN TO USE — invoke this tool for ANY of these intents:

  • "generate an image of...", "create a picture of...", "make me an image...", "add an image ...",
  • "I need an image for my slide about..."
  • "draw", "illustrate", "visualize", "design an image", "create a visual"
  • "can you add a photo of...", "make a graphic showing..."
  • "infographic", "custom diagram", "flowchart", "data visualization", "chart illustration" — these are VISUAL concepts that require AI image generation, not text layouts
  • User asks for a slide "with an infographic" or "with a diagram" — this means generate the visual FIRST
  • User describes a specific scene, concept, or visual they want created from scratch
  • User wants a particular image that stock photos cannot provide
  • "I want a slide with an image of [specific thing]" — if the described image is specific enough that a stock photo won't match, use this tool

  DO NOT USE when:

  • The user wants a generic stock photo (handled automatically by create_slides via image_caption)
  • The user uploaded a file with images (use upload_files instead)

  HOW IT WORKS:

  1. Call generate_image with a detailed prompt describing the desired image
  2. The image is generated and returned with a URL
  3. Present the image URL to the user and ask if they want to use it
  4. WAIT for the user to confirm — do NOT call create_slides until the user accepts
  5. When the user accepts, call create_slides with:
     • custom_image_url set to the generated image URL
     • presentation_id from the generate_image response (to add to the same deck)
     • An image layout: "image_full_width", "image_left_text_right", or "title_and_two_columns"

  CRITICAL RULES:

  • If a presentation already exists in this conversation, you MUST pass the presentation_id
  • Write a rich, detailed prompt — the more specific, the better the image
  • After receiving the image, ALWAYS show it to the user and ask for confirmation before proceeding
  • If the user rejects, offer to regenerate with a modified prompt

  PROMPT WRITING TIPS (include in your prompt to get better results):

  • Subject: What is the main focus of the image?
  • Setting/Environment: Where is the scene taking place?
  • Style: Photorealistic, illustration, flat design, etc.
  • Mood/Lighting: Warm, cool, dramatic, soft, etc.
  • Composition: Close-up, wide angle, aerial view, etc.
  • Color palette: Specific colors or general feel
  • NEVER ask for text or words in the image — they will not render correctly