add_slide
ChatGPTAdds one slide at the end and returns its number. layout is a layout name as returned by list_layouts (for example Title Only, Title and Content, Two Content, Blank; names follow the Office language). Without layout, PowerPoint's default layout is used (often a Title Slide).
add_text_box
ChatGPTPlaces a text box with regular PowerPoint text. Use it for text without math.
delete_shape
ChatGPTDeletes one shape (also used to clear empty placeholders). shapeId is an id from list_slides (slide/shape). Undo in PowerPoint restores it.
figure_build
ChatGPTTypesets code (only the tikzpicture for .tikz, only the tblr environment for .tblr) with preamble (\usepackage lines, etc.) and returns the result image. With a source image it also returns an overlay (red = only in the source, blue = only in the build). Also returns the request-table tally and, for tables, a cell check and cells per row.
figure_guide
ChatGPTReturns the procedure for tikzify (figures to TikZ) and tablify (tables to tabularray). Read it once before making a figure or table. topic reads a reference (requests, pitfalls, sources, figure-types, handdrawn, tabularray), the template list (templates), or a template by name. The guide is written in Japanese.
figure_measure
ChatGPTMeasures line positions, circles, curves, text blocks, and fill darkness in the coordinate system of the last tikz_grid. args are measure.py arguments (e.g. ["--row","2.0","--circle","8.4,3.2,1.9,2.5"]).
figure_place
ChatGPTPlaces the last built figure or table as SVG on a slide (or in Word). With replaceShapeId it replaces that shape (the source image or a previous version) at the same position and width. Without widthPt it uses the source image width, or else the built size. Refuses while the request table still has 未 (open) items (force overrides).
figure_requests
ChatGPTRewrites the whole request table (set state to 済 when done; 不可 with the reason in how when not possible). Add rows for new requests.
figure_source_from_slide
ChatGPTExtracts an image pasted on a slide (photo or screenshot) as the source for this work. Its position and size are remembered so figure_place can replace it via replaceShapeId.
figure_source_text
ChatGPTPasses a table as CSV / TSV / Markdown / HTML text (for example, a table the user pasted).
figure_start
ChatGPTCreates a workspace and records the user's requests. userWords is the user's own words, verbatim. requests is one request per item (state: 済 done / 一部 partial / 不可 not possible / 未 open). Mark wishes inferred from the document or conversation with "(推定)" in request.
insert_graph
ChatGPTPlots graphs of functions, curves, surfaces, points or vectors given by formulas (2D or 3D) and places them on a slide (or in Word) as a sharp vector figure. Diagrams that are not plots of formulas are drawn with the figure_* tools. The user can later select the graph in the MathPoint pane and edit the same expressions. Returns the placed size, the axis ranges used, and an image of the result: look at it and fix ranges or expressions if needed (update_graph). One expression per item. Plain (y = sin(x)^2) or LaTeX (y = \sin^2 x) both work; implicit multiplication is fine (2x, a x^2). - 2D curves: y = f(x) · x = g(y) · r = f(θ) · (x(t), y(t)) · implicit equations x^2 + y^2 = 4, x y = 1 - Shaded regions: inequalities y > x^2, x^2 + y^2 < 9, 1 < x^2 + y^2 < 4 (a strict < or > draws a dashed edge). Area under a curve: 0 ≤ y ≤ f(x) {1 ≤ x ≤ 3}; between curves: x^2 ≤ y ≤ x {0 ≤ x ≤ 1} - Restrictions: append {…} conditions, as many as needed: y = x^2 {0 ≤ x ≤ 2}, (cos t, sin t) {0 ≤ t ≤ π} - Piecewise: y = {x < 0: -x, x^2} - Points: A = (1, 2) (labelled; usable later as A, A.x), lists [(1, 2), (2, 4)] (data); y ~ m x + b fits m and b to every point in the graph - Lists: a = [1, 2, 3] with y = a x^2 draws one curve per value - Shapes: vector(1, 2) (from the origin), vector(A, B), segment(A, B), polygon(A, B, C), circle(A, 2) - Functions: f(x) = x^3 - 3x is drawn and can be used later: f(x - 1), f'(x), f''(x); also \frac{d}{dx}(…), \int_{0}^{x} g(t)\,dt, \sum_{n=1}^{10} … - 3D: z = f(x, y) · x = f(y, z) · z = f(r, θ) (cylindrical) · (x(u, v), y(u, v), z(u, v)) parametric surface (u, v default 0…2π and 0…π with trig) · implicit surfaces x^2 + y^2 + z^2 = 4 · (x(t), y(t), z(t)) · P = (1, 2, 3) · vector, segment, polygon with 3D points - Parameters: a separate item a = 2 defines a Functions: sin cos tan sec csc cot, arcsin…, sinh…, exp, ln, log, sqrt, abs, floor, ceil, sign, min, max; constants π, e.
insert_image
ChatGPTTakes a PNG / JPEG as base64 and places it on a slide (for AI clients that can read local files). Suited to photos, screen captures, and logos. A figure placed this way can later be redrawn as a TeX figure (figure_source_from_slide).
insert_math
ChatGPTTypesets LaTeX with TeX and places it on the given slide as SVG. mode: display (set apart), inline (in running text), text (a whole phrase mixing text and $...$).
list_layouts
ChatGPTLists the layout names in this presentation's slide master, for add_slide's layout.
list_slides
ChatGPTReturns every slide of the open presentation and its shapes (id, type, position, text, MathPoint tags).
move_shape
ChatGPTChanges a shape's position (left, top) and size (width, height) in pt. Omitted values stay as they are. Use it, for example, to move a title placeholder that overlaps a figure. TeX math, figures, and tables keep their aspect ratio: changing only width also changes height.
render_graph
ChatGPTPlots function graphs like insert_graph and returns only the image. Does not change any slide or document, and works without PowerPoint or Word open (explaining in chat, use from a phone). One expression per item. Plain (y = sin(x)^2) or LaTeX (y = \sin^2 x) both work; implicit multiplication is fine (2x, a x^2). - 2D curves: y = f(x) · x = g(y) · r = f(θ) · (x(t), y(t)) · implicit equations x^2 + y^2 = 4, x y = 1 - Shaded regions: inequalities y > x^2, x^2 + y^2 < 9, 1 < x^2 + y^2 < 4 (a strict < or > draws a dashed edge). Area under a curve: 0 ≤ y ≤ f(x) {1 ≤ x ≤ 3}; between curves: x^2 ≤ y ≤ x {0 ≤ x ≤ 1} - Restrictions: append {…} conditions, as many as needed: y = x^2 {0 ≤ x ≤ 2}, (cos t, sin t) {0 ≤ t ≤ π} - Piecewise: y = {x < 0: -x, x^2} - Points: A = (1, 2) (labelled; usable later as A, A.x), lists [(1, 2), (2, 4)] (data); y ~ m x + b fits m and b to every point in the graph - Lists: a = [1, 2, 3] with y = a x^2 draws one curve per value - Shapes: vector(1, 2) (from the origin), vector(A, B), segment(A, B), polygon(A, B, C), circle(A, 2) - Functions: f(x) = x^3 - 3x is drawn and can be used later: f(x - 1), f'(x), f''(x); also \frac{d}{dx}(…), \int_{0}^{x} g(t)\,dt, \sum_{n=1}^{10} … - 3D: z = f(x, y) · x = f(y, z) · z = f(r, θ) (cylindrical) · (x(u, v), y(u, v), z(u, v)) parametric surface (u, v default 0…2π and 0…π with trig) · implicit surfaces x^2 + y^2 + z^2 = 4 · (x(t), y(t), z(t)) · P = (1, 2, 3) · vector, segment, polygon with 3D points - Parameters: a separate item a = 2 defines a Functions: sin cos tan sec csc cot, arcsin…, sinh…, exp, ln, log, sqrt, abs, floor, ceil, sign, min, max; constants π, e.
render_math
ChatGPTTypesets LaTeX with TeX and returns a PNG image. Does not change any slide or document, and works without PowerPoint or Word open (checking a formula, explaining in chat, use from a phone). mode: display (set apart), inline (in running text), text (a whole phrase mixing text and $...$).
select_file
ChatGPTMakes the named open file (PowerPoint or Word) the target of the following tool calls. Intended for when several files are open (listed in usage.files). Matching ignores the extension and case, and accepts part of the name.
set_text
ChatGPTReplaces all text of an existing shape (text box, placeholder, or shape). shapeId is an id from list_slides (slide/shape).
table_draft
ChatGPTBuilds a booktabs-style draft .tblr and cells.json from the source data. Mostly numeric columns become decimal-aligned S columns. When the source is an image, rows / cols build an empty skeleton.
tikz_grid
ChatGPTReturns the source image with a coordinate grid (red lines every 1 cm, origin bottom-left, y up). widthCm is the target width on the slide (cm). Use clean for photos. frame / mask / zoom work as in grid.py (cm).
update_graph
ChatGPTRedraws a graph MathPoint inserted (MP_KIND graph; its expressions and view are in tags.MP_GRAPH). Omitted fields keep their current values: pass only what changes (for example new expressions, a new x range, or a 3D view angle). Keeps the position and the width (including a manual resize) unless width is given. One expression per item. Plain (y = sin(x)^2) or LaTeX (y = \sin^2 x) both work; implicit multiplication is fine (2x, a x^2). - 2D curves: y = f(x) · x = g(y) · r = f(θ) · (x(t), y(t)) · implicit equations x^2 + y^2 = 4, x y = 1 - Shaded regions: inequalities y > x^2, x^2 + y^2 < 9, 1 < x^2 + y^2 < 4 (a strict < or > draws a dashed edge). Area under a curve: 0 ≤ y ≤ f(x) {1 ≤ x ≤ 3}; between curves: x^2 ≤ y ≤ x {0 ≤ x ≤ 1} - Restrictions: append {…} conditions, as many as needed: y = x^2 {0 ≤ x ≤ 2}, (cos t, sin t) {0 ≤ t ≤ π} - Piecewise: y = {x < 0: -x, x^2} - Points: A = (1, 2) (labelled; usable later as A, A.x), lists [(1, 2), (2, 4)] (data); y ~ m x + b fits m and b to every point in the graph - Lists: a = [1, 2, 3] with y = a x^2 draws one curve per value - Shapes: vector(1, 2) (from the origin), vector(A, B), segment(A, B), polygon(A, B, C), circle(A, 2) - Functions: f(x) = x^3 - 3x is drawn and can be used later: f(x - 1), f'(x), f''(x); also \frac{d}{dx}(…), \int_{0}^{x} g(t)\,dt, \sum_{n=1}^{10} … - 3D: z = f(x, y) · x = f(y, z) · z = f(r, θ) (cylindrical) · (x(u, v), y(u, v), z(u, v)) parametric surface (u, v default 0…2π and 0…π with trig) · implicit surfaces x^2 + y^2 + z^2 = 4 · (x(t), y(t), z(t)) · P = (1, 2, 3) · vector, segment, polygon with 3D points - Parameters: a separate item a = 2 defines a Functions: sin cos tan sec csc cot, arcsin…, sinh…, exp, ln, log, sqrt, abs, floor, ceil, sign, min, max; constants π, e.
update_math
ChatGPTRe-typesets a shape MathPoint inserted. Omitted fields keep their current values. Keeps the top-left position, and keeps any manual resize as a scale factor.
usage
ChatGPTHow many items were typeset with TeX this month and the free quota left. apps lists the apps with the MathPoint task pane open (PowerPoint / Word); files lists the open files and which one is current for each app.
word_delete
ChatGPTDeletes one MathPoint formula, figure, or table (id) or one paragraph (paragraph). Undo in Word restores it.
word_insert_image
ChatGPTTakes a PNG / JPEG as base64 and places it in a new centered paragraph after paragraph (photos, screen captures, logos).
word_insert_math
ChatGPTTypesets LaTeX with TeX and inserts it into the Word document as an image. where: newParagraphAfter (default; a new centered paragraph after paragraph = display math), replaceText (replace find inside paragraph with the formula = inline math), afterText (insert right after find), cursor (the user's cursor). Inline math matches the body text size (set fontSize to the body size) and sits on the baseline.
word_insert_text
ChatGPTAdds one paragraph of regular Word text (heading, body, list item; text without math). style is a Word style name (e.g. "Heading 1"; names follow the Office language). Markers such as ⟦1⟧ in the text can later be replaced with a formula by word_insert_math (where=replaceText).
word_read
ChatGPTReturns the paragraphs of the open Word document (number, style, first 300 characters) and the math, figures, and tables MathPoint inserted (id and LaTeX).
word_set_text
ChatGPTReplaces all text of a paragraph (math and images in it are removed too).
word_update_math
ChatGPTRe-typesets a formula MathPoint inserted in Word, replacing it in place. id is mathpoint[].id from word_read (mp:…). Omitted fields keep their current values.