Project Book
Read and edit the saved history of a Glowbom project.
Loading page content...
Project Book turns saved prototypes and builds into a readable project history. The backend, generation APIs, and portable records are open source. The reader and editing screens described below belong to the separate Desktop client, not the minimal public Agent + Buzz interface.
Open a project in Glowbom Desktop and select the small book icon beside the chat controls, in Result, or in the full workspace. Opening the Book gathers saved records without calling a model or changing project code.
Select an entry to read its story and view any saved illustrations. Behind this story shows the captured request, contributor, recorded models, changed files, and run outline when available. Older records may have fewer details.
After a successful prototype or supported build, an available Book model can draft a story and a vector drawing from the saved work. The drawing is an AI interpretation, not a verified screenshot of the running app. View drawing opens it; the window icon opens an editable copy in Magic Window.
Use the pencil to edit the title, story, and captions. Add PNG, JPEG, or WebP illustrations with the image icon. Each save creates a revision. New records and retries preserve existing stories, drawings, and owner edits.
If content is missing, choose Drawing and text, Drawing only, or Text only, then Retry selected. An untouched older entry may offer Write story. Keep Glowbom open while generation runs. Use the speaker to listen with your saved chat speech voice.
Open Account > Settings > Project Book > Model. This model generates stories and drawings independently of Build and Chat. Automatic tries an eligible build model, then your Chat model. Compatible connected OpenCode and Codex models are available; build-only agents such as Cursor use a separate eligible Book model.
An entry's Using > Change control overrides the model for that entry without changing the default. Changing models leaves saved content intact.
Settings also offers writing voices and a Custom style prompt. These apply to new stories and drawings. Try another voice previews a rewritten story; review it and select Save entry to keep a new revision, or cancel. The preview leaves the drawing intact.
The Book stays inside the project:
project-book/
README.md
book.json
intent.md
current-result.md
decisions.md
next.md
PROJECT_BRIEF.md
assets/
history/
<entry-id>/<revision>/
entry.md
source.json
sketch.json
ready
entry.md holds the editable story, source.json records its evidence, and
sketch.json holds vector drawings. The latest revision with a ready marker
is current. The overview notes are owner starting points; PROJECT_BRIEF.md
initially points to them. Keep the project's original history/ and
.glowbom/prototypes/ records when moving it.
Prototype requests to POST /chat/stream can include a carousel object with
version: 1 and up to ten cards. Each card has a stable id, a role, a
title, and optional instructions. Roles are sketch, person, or
reference, and each carousel needs at least one sketch. Titles allow 200
Unicode characters, and instructions allow 32,000.
All cards contribute to the same prototype. A personalized image can appear
inside a layout described by another card.
An optional attachmentPath refers to an image from the existing upload API.
A sketch can include an inputSketch drawing object. A sketch with
personalization: true uses only its own image as a generation reference.
This requires an image and excludes personIds. Without personalization or
selected people, a sketch receives no person reference from other cards or
previous images. Requests without carousel keep the existing single-image
behavior.
The saved prototype version contains input-carousel.json beside
request.json. The request records inputCarouselPath: "input-carousel.json".
The saved card replaces upload paths with relative paths inside inputs/.
Drawing objects are separate JSON files named by inputSketchPath:
{
"version": 1,
"cards": [
{
"id": "scene-1",
"role": "sketch",
"title": "First scene",
"instructions": "Show the supplied subject beside the sea.",
"attachmentPath": "inputs/card-scene-1.png",
"inputSketchPath": "inputs/card-scene-1-sketch.json",
"personalization": true
}
]
}
The API also supports selected groups through sketch personIds and photo
cards with matching personId labels. Labels match exactly after trimming
outer spaces. Multiple photo cards can share a label. The generator receives
a numbered sheet containing only the selected photos. Person-card photos stay
out of the prototype planning request. Original inputs remain in the saved
version and are not copied into public prototype assets.
Generated image descriptions use glowbomimages:[card:scene-1] description to
identify their sketch. Missing or invalid card tags skip that image. The
existing four-image maximum and approved credit limit still apply.
The iOS and Vision Pro companion can read Books for shared projects; editing stays on Desktop. Reading or editing a Book does not rebuild software or publish it. Generated stories and drawings need review. Images are limited to 8 MiB and 32 megapixels, with up to 32 images per entry and 1000 entries per Book. Older releases may not include this reader.