Your ideas, connected.

Try a small project

• Choose a project and capture a few ideas or tasks. Short, unfinished thoughts are a useful starting point.
• Open a note and add the detail you want to preserve. Give it a title that will make sense when you return.
• Place related notes together on the canvas and link them. A visual grouping helps you navigate; a link makes a relationship explicit.
• Select a note, ask a question about your material, or request a synthesis. Start with a small scope so you can compare the result with its sources.

Keep the source material

A synthesis is another view of your notes. Keep the original evidence and your own interpretation available, and return to them when an AI answer seems incomplete. You can keep capturing and editing without organizing everything first.

iPhone

The smaller screen uses layered views: the canvas, list or tool drawer, and note editor share the available space. Use the toolbar to reach search, capture, and canvas controls. Close the current editor or panel to return to the graph.

iPad & Mac

Larger screens place the list beside the canvas and provide docked or floating editor arrangements. Pointer and keyboard controls complement direct manipulation. The exact arrangement depends on the device and window size.

Tools & help

The tool selector brings together chat, Import Studio, DataJars, Knowledge Browser, and Prompt Architect. Glob Wiki opens this website in your browser; prompt editing stays in the app. Soft panels and restrained color accents help keep the graph visible behind your current task.

Move from detail to pattern

Pan to explore the workspace and zoom to move between a close view of a few notes and the shape of a larger network. Select a node to work with it, drag it to rearrange the graph, and use linking controls to connect related ideas. Selection tools and keyboard navigation vary by platform.

Groups, trees & networks

A parent-and-child tree describes a hierarchy. A connected network can include relationships that cross that hierarchy. This difference matters when choosing a synthesis: a subtree follows descendants, while a component follows connections. Nearby notes are not automatically the same as linked notes.

Summary views

Where available, switch between node and summary resolution to inspect the meaning of a group instead of every individual note. Project and cluster summaries help you step back; the note editor lets you return to the detail.

Develop a note

Open a node to edit its title and body. Note drafts save automatically as you work. If you leave the title blank, Glob can derive a display title from the body. Empty notes can be cleaned up when the editor closes, so add content to a note you want to keep.

Read a synthesis beside its source

Summary blocks appear separately from the note body. Generating a summary does not replace your writing. When a synthesis offers variants, you can choose the version that is most useful. Human-edited, pinned summaries are protected from automatic replacement.

Collect first, organize when useful

Use the capture bar for quick thoughts and tasks. The list gives you focused views such as tasks, ideas, and clusters. Promote useful material into the current project when you are ready to connect it with other notes.

Find material again

Toolbar search adapts to where you are working: it can filter the active list or open canvas search when the list is closed. Canvas search can bring a matching node into view. Check the current view and project when the result is not where you expected.

Route a capture to a DataJar

Pin a jar from the capture menu to direct captures into it. Dot-prefix commands can address a known jar for capture or search. The jar’s slug is its short address; choose a memorable one when creating the jar.

Ask a question with a clear target

Name the idea, project, or source you want to discuss. For example: “What tensions appear between these notes?” or “Which evidence supports this proposal?” A general chat question can take a fast path without local knowledge context. Do not assume every answer has read your graph.

What contextual chat can use

For contextual requests, Glob assembles relevant material at send time. This can include the selected node, connected or nearby notes, parent and child summaries, project context, recent captures, and retrieved document or DataJar excerpts. Conversation summaries can help carry context between turns.

Context is selective

Retrieval uses bounded packets of relevant material. It is not a complete copy of your library, and large sources may be condensed or omitted. Narrow the question or identify a source explicitly when important evidence is missing. Hosted requests send the selected context for processing; local storage alone does not mean an AI request stays on the device.

The built-in layers

• System sets the general synthesis behavior and rules.
• Scopes describe the task for a node, subtree, component, or project.
• Styles describe how to interpret the material, such as structural analysis, perspective, or simplification.
• Features provide instructions for specialized operations such as Lookup, Brainstorm, Clusters, Network Chat, and document import.

Custom prompts

A custom prompt supplies its own system and task instructions for a synthesis. An empty system or task field falls back to the corresponding default. The custom path does not append the normal style layer. Scope, target length, evidence, framing context, and output requirements still accompany the request.

Save & reset

Use Done to save changes in the standalone editor. The embedded Prompt Architect saves when it disappears. Reset to Default restores the selected built-in layer; it is not a reset of every prompt. Custom prompts can be named and managed separately.

Make changes you can evaluate

Keep instructions short and specific. Describe the desired output, how to handle missing evidence, and what distinctions matter. Try a change on familiar notes, compare the result with the source, and adjust one layer at a time. Prompt preferences use iCloud key-value storage where available, separately from account-based content sync.

Bring in a source

Supported extraction paths include plain text, Markdown, CSV and TSV, HTML, DOCX, PDFs, and images. PDFs can use selectable text or OCR for scanned pages; images use OCR where supported. Results depend on the source quality and the extraction capabilities available on your device.

Review before adding knowledge

Glob retains the imported source locally, extracts text, breaks it into chunks, and checks for duplicates. Review the extracted material, especially tables and OCR text. Preflight warnings can help identify a large or difficult file before you continue.

Use only what you need

Turn reviewed chunks or selected text into snippets, tasks, canvas nodes, clusters, networks, or DataJar entries through the available import actions. Document Q&A retrieves relevant chunks instead of assuming the entire document fits into an AI request. Keep source references when you need to verify an answer later.

Give a jar a purpose

A jar has a name and slug, visual identity, and a definition that can include fields, retrieval policy, privacy mode, and allowed actions. These settings describe how entries should be organized and used. Available field controls can differ by build.

Add and retrieve entries

Capture directly into a pinned jar, route a capture with its dot-prefix command, or add reviewed material from Import Studio. Jar-scoped search and contextual AI use relevant entries and summaries to keep retrieval focused.

Review AI proposals

Chat can produce proposed DataJar entries. In the proposal review flow, inspect the content and explicitly apply or dismiss it. An AI suggestion is not itself a saved entry. Do not rely on a permanent proposal history unless your build explicitly provides one.

Browse, then bring material into a project

Filter by source to find records, inspect the material, and import a selected record as a canvas note. On Mac the browser can appear as a floating panel; it is also available inside the shared tools panel. Older source labels may appear on historical records.

Choose the right surface

Use Import Studio to prepare a document, a DataJar to maintain a focused collection, and Knowledge Browser to inspect the shared knowledge layer. A record being available for browsing does not mean every chat automatically receives it.

Choose a scope

• Node focuses on the selected note. If its body is empty, direct neighbors can supply context.
• Subtree includes a root and its descendants in the structural tree.
• Component follows the connected network containing the selected note, including connections beyond a single branch.
• Project gathers non-trashed notes in the active project for a broader view.

How source material becomes a result

Glob collects titles and bodies and removes embedded legacy summary blocks to avoid repeatedly summarizing summaries. Collection uses a text budget, so a broad scope does not guarantee that every note appears in the request. A local extraction can prepare the evidence and provide a fallback; an available AI route can refine it. Some operations stream results as they arrive.

When AI cannot run

Eligibility depends on the operation, source material, connectivity, account access, and available AI route. The ordinary note synthesis pipeline can retain a local result when AI is unavailable or fails. Specialized operations, including structured project summaries, may instead report that AI is required. There is no fixed completion-time guarantee.

Freshness & deeper synthesis

Glob compares source fingerprints to reuse unchanged results and detect stale work. A newer request can cancel work for the same result slot. Deeper synthesis can work through a hierarchy to build a broader interpretation. After changing important source material, review whether the displayed synthesis still reflects what you mean.

Separate from your writing

Results are stored by project, target node, scope, and mode. Regenerating updates that slot rather than adding duplicate text to your note. Blocks can contain variants and structured results; pinned human edits receive protection from automatic updates. Summary blocks are saved locally and mirrored through iCloud key-value storage where available, separately from the main content sync transport.

Set up each device

Follow the account sign-in flow offered by your build. When sync is enabled, use the same account and sync passphrase on each device. The passphrase is kept in the device’s Keychain and is not sent to the server. The app’s unlock flow explains that the passphrase cannot be recovered.

Know what sync covers

The content sync path includes knowledge records, links, collections, and DataJar definitions. CloudKit database mirroring is not the current main content transport. Prompt overrides and summary blocks still use a separate iCloud key-value path, so their availability across devices can differ from account-synced notes.

If something is missing

Check the signed-in account, network connection, and sync or passphrase state shown by the app. Check that both devices are using compatible builds. Imported source files and other local supporting data should not be assumed to transfer merely because a record appears on another device.

AI access

Hosted Glob API is the normal cloud AI path. Personal provider keys are not required for that account-based setup. On-device AI is used only where the feature, device, and implementation support it; it is not a universal offline substitute. Consult your account and the current download information for access and allowances.

Older documentation

Older material may describe Glob Works as a companion workflow app, direct provider-key setup as the standard path, or CloudKit as the main database sync service. Those descriptions are not the current setup guide. This wiki does not require Glob Works or promise its workflow automation features.

Work in progress

Specialized DataJar controls, proposal history, and system integrations should be treated as available only when your build exposes them. Use the app’s status and error messages to distinguish a local result, an AI result, and an operation that could not complete.