Thinking with your notes
Prompt Architect
Edit the system, scope, style and feature instructions that shape Glob's generated results.
Prompt Architect lets you change how Glob interprets material without rewriting the material itself. Use it when the output needs a different emphasis: shorter explanations, clearer distinctions, or more explicit handling of missing evidence.
Open the list panel's conversation/tools view and choose Prompt Architect. You can also use Edit Summary Prompts in Settings → AI Summarization for the standalone editor.
Choose the layer that owns the change
The standalone editor groups its choices under System, Scopes, Styles, Features, Wiki, and Custom. The embedded Prompt Architect uses a horizontal row of choices; scroll it to reach the later ones.
| Layer | What it changes |
|---|---|
| System | The general synthesis behaviour used across scopes. |
| Scopes | The task for Node, Subtree, Component, or Project. |
| Styles | Synthesis, Structural, Perspective, or Simplification. |
| Features | Instructions for a particular tool rather than ordinary synthesis. |
Feature choices are Lookup, Brainstorm, Clusters, Network Chat, Document Import, and Document Q&A. Editing Document Q&A, for example, changes that question-answering instruction; it does not alter the text extracted from an imported document.
A small dot beside a built-in choice in the standalone sidebar means it has a saved override.
How built-in synthesis is assembled
A normal synthesis combines the System instruction, the selected scope's task instruction and the selected style instruction. Glob also supplies the scope description, target length, evidence, framing context and output requirements.
That means a shorter instruction can work better than repeating the whole task. For example, add “Separate observations from interpretations. If the notes disagree, preserve the disagreement.” to the appropriate instruction, then test it on notes whose contents you know.
The instruction affects operations that use that layer. It does not force every other tool to use the same prompt, add missing source material to a request, or switch the AI provider.
Save a change
In the standalone Edit Prompts editor, choose Done to save. Do this before closing its window. Closing a standalone window is not a reliable save action in this build.
The embedded Prompt Architect saves when you leave that view. Switching to another tool or going back lets it save the edited layers.
Saved prompts affect later requests. Existing summaries are not rewritten just because you edited an instruction. Generate or refresh a result to evaluate the change.
Reset one built-in layer
Select the layer and choose Reset to Default. The reset applies to that selected System, scope, style or feature instruction. It does not reset every prompt or delete your custom prompts.
If a result became less useful after an edit, reset the layer you changed and repeat the same small example. Keep the evidence and other layers the same so you can tell what the instruction changed.
Custom prompts
The standalone editor's Custom section includes New Prompt…. A custom prompt has Name, System Instruction, and Task Instruction. Use Done after editing it.
For a synthesis request that uses a custom prompt, its System Instruction replaces the built-in one. An empty custom System Instruction falls back to the saved System layer. An empty Task Instruction falls back to the scope's task instruction.
The custom path does not append the normal style instruction. Scope, target length, evidence, framing context and output requirements still accompany the request. A custom instruction therefore does not give you an unbounded input or remove the result's expected format.
Use Delete Prompt to remove a custom prompt. Creating or naming one does not by itself select it for the hexagon's branch action; that action uses its own default configuration in this Mac build.
Keep prompt experiments small
Choose familiar source notes and change one layer at a time. Ask for a concrete result, such as a concise explanation with unresolved questions kept separate. Broad claims such as “always be correct” do not give the model missing evidence.
If an answer omits a source, changing the instruction may not solve the problem. Chat and synthesis use selective context. Narrow the question or source scope first; see Ask Glob and Synthesis.
Prompt preferences are saved separately from notes and use Apple's preference synchronisation. Do not expect a prompt edit to arrive in the web app through Glob account note sync.
