Inputs & Context

The Brainstorm Reactor's analysis is only as good as the context you provide.

A major flaw in generic AI chatbots is the "empty text box" problem. Faced with an open input field, users supply insufficient, ambiguous, or poorly constrained instructions. The result is generic, unusable output.

The Brainstorm Reactor counters this failure pattern by structuring the input area into a dedicated 4-Zone Smart Input Architecture, combining proactive context anchors, ambient intent triage, and robust multi-modal ingestion.

1. The 4-Zone Input Architecture

The user input area (ReactorInputArea) is organized into four distinct, vertically coordinated functional zones:

┌─────────────────────────────────────────────────────────────┐
│ ZONE 1: Active Context Anchor                               │
│ [ 🧠 6-3-5 Brainwriting ✕ ] [ Step 1 / 4 ] [ ✓ Briefing ]   │
├─────────────────────────────────────────────────────────────┤
│ ZONE 2: Smart Assistant Shelf (Single-Slot Priority Shelf)  │
│ [ ✨ Rezept: Nordic Slow Cinema ]                           │
├─────────────────────────────────────────────────────────────┤
│ ZONE 3: Multi-Line Prompt Canvas                            │
│ "Ich suche eine Idee für einen Kurzfilm..."                 │
│                                                             │
├─────────────────────────────────────────────────────────────┤
│ ZONE 4: Action Toolbar & Model Cockpit                      │
│ [+]                         [GPT-5.6 Luna ▾]       [ ▶ ]    │
└─────────────────────────────────────────────────────────────┘

Zone 1: Active Context Anchor

The Active Context Anchor sits at the top of the input card and displays persistent badges representing your active session state:

Zone 2: Smart Assistant Shelf

The Smart Assistant Shelf is a single-slot proactive intelligence bar. It ensures cognitive quietness by rendering at most one compact suggestion row at any given moment, governed by a strict priority order:

1. Priority 1: Conductor Ambient Triage & Lexical Match
   • Lexical Method Match (Direct Mode): Typing a method name or alias (e.g. scamper, walt disney, pre-mortem, swot) immediately shows a silver pill (e.g. [Methode: SCAMPER]). Clicking it activates the method in Zone 1.
   • Lexical Recipe Match (Workflow Mode): Typing a recipe name or core title (e.g. pitch architect, hegelian forge, triz, neumarkt) immediately shows a lavender pill (e.g. [Rezept: Pitch Architect]). Clicking it loads the workflow and preserves your typed prompt.
   • Catalytic Recommendations (Problem Prompts): Typing a substantive challenge without specific method or recipe names surfaces an ambient power search ([Direkt: 3 Methoden empfehlen] in Direct Mode or [Workflows: 3 Rezepte empfehlen] in Workflow Mode) without interfering with specific title lookups.
   • Visual Engine Intent: Prompting for renders, mockups, or visual styles (e.g. rendering, mockup, visualisierung, bild generieren) immediately surfaces [Bild generieren]. Clicking routes the prompt directly into the Director's Studio console.
   • Search Intent & Link Grounding: Generic search triggers operate in Auto Mode without intrusive popup badges, allowing the model to autonomously query external sources on demand. When external URLs or web links are included in the prompt, the shelf renders a targeted [Link-Recherche aktivieren] action. When substantive briefs contain URLs, the primary recommendation is displayed first, followed by the link research chip side-by-side.
2. Priority 2: Method Recommendations
   • Displays top-3 blue method pills suggested by the AI Dispatcher.
3. Priority 3: Recipe Recommendations (Dual-Tier 8-Slot Engine)
   • Displays top-3 purple recipe suggestions with a quick [+ 5 mehr] extension button and a one-click dismissal [✕] control.
   • Clicking [+ 5 mehr] automatically opens the Left Sidebar in expanded view, rendering all 8 epistemically categorized recommendations (Tier 1: Direct Convergence; Tier 2: Perspectives & Stresstests), while hiding the input shelf pills to eliminate visual duplication.
4. Priority 4: Golden Upgrade Chip
   • Suggests model set upgrades ([⚡ LLM Set Upgrade empfohlen]) when complex analytical prompts are typed while using budget models.
5. Priority 5: Proactive Privacy Nudge
   • Automatically detects sensitive personal identifiable information (PII)—including German/EU IBANs, email addresses, and financial keywords—and renders [🛡️ EU Privacy Gateway aktivieren] to route prompts through the EU privacy gateway.
6. Priority 6: Smart Context Output Nudges
   • Renders context-aware export actions ([📄 PDF Export], [📊 Diagramm], [🌐 Web Preview]).

[!NOTE] Smart Paste & Document Triage: Pasting long text (> 2.500 chars) automatically converts it into a neutral attached document (Notiz.txt in DE / Note.txt in EN). The recommendation engine and method matcher analyze up to 5.000 characters of attached text files upon user click, strictly adhering to the EU Privacy Gate (zero background LLM calls on paste).

Zone 3: Multi-Line Prompt Canvas

The main textarea supports responsive multi-line editing and full-bleed focus:

• Adaptive Height Bounds: Automatically expands with text volume up to 7 lines on Desktop and 6 lines on Mobile before engaging internal scrolling.
• Fullscreen Desktop Expansion: Typing 3 or more lines reveals a Maximize button (Maximize2) in the top-right corner. Expanding transforms the input field into a full-height console overlay across the central chat canvas—ideal for crafting comprehensive project briefs or editing extensive prompt structures.
• Effortless Collapse: Pressing Esc, clicking Minimize (Minimize2), or dispatching the prompt (Enter) smoothly restores the compact resting state.
• Unified Dispatch Trigger (Enter): Following modern prompt and chat ergonomics, pressing Enter (Return) dispatches your prompt across all execution modes. To insert a clean newline (soft return), press Shift+Enter. For power users, ⌘+Enter (Mac) or Ctrl+Enter (Win/Linux) remains fully supported.
• Cognitive Ergonomics: Rotating placeholder suggestions cycle at a calm 9-second cadence with 1-second cross-fades, paired with an optically softened caret tone to prevent visual fatigue on dark interfaces.
• Text Preservation Guarantee: Typed input is 100% preserved during mode transitions, recipe selections, and background state re-renders.

Zone 4: Action Toolbar & Unified Model Cockpit

The bottom action bar provides direct access to multi-modal tools, live audio transcription, and the Unified Model & Execution Cockpit:

• Attachment & Tools Menu (+): Upload external files, generate image visuals, or start live audio transcription with 3 specialized modes (Live Insights, Wortprotokoll, Social Media).
• Active Audio Indicator: During active voice recording, an unobtrusive pulsing red stop button appears directly in the prompt bar for quick completion.
• Unified Model & Execution Cockpit: The primary model selector bridges Direct Mode and Workflow Mode within a single unified control, eliminating mode fragmentation:
  • Direct Mode (Resting State): Displays a focused, clean trigger badge ([● Balanced] or [● Model Name]), protecting focus without distracting credit numbers on the outside.
  • Recipe / Workflow Mode: Seamlessly morphs into the Recipe Execution Cockpit, displaying the active synthesis engine alongside an automatic Mono badge when all three slots share the same model.
  • Dynamic Workflow Credit Calculation: Instead of static estimates, the cockpit sums the real-time credit costs across all planned recipe steps by resolving the exact model assigned to each step's execution slot (Synthesis, Logic, Search), displayed as the single authoritative credit count in the popover header.
• The Cockpit Popover (Zero-Scroll Ergonomic Design):
  • Unified Macro-Tier Cockpit: Focuses on the 3 core tiers (Speed | Balanced | Deep, plus Ultra when unlocked). Switching tiers instantly updates slot allocations and expected credits.
  • Datenschutz & Compliance Toggles: Provides dedicated switches for In der EU gehostet (guaranteeing 100% European server residency) and EU PII-Shield (active pre-inference redaction via EU PII-Shield).
  • Planned Model Slots (Geplante Modell-Slots): In Workflow Mode, the popover displays the functional slots (Synthese, Logik, Recherche) with 1-click mono synchronization and direct access to customize individual slots (Slots anpassen ↗).
  • Full Expert Catalog: Choosing a specific model outside the standard tiers is accessible via the expandable catalog drawer (Spezifisches Modell wählen (27)…), with a 1-click reset to auto-tiers.
  • Web Search Integration: Grounded web search is triggered directly from the primary toolbar via the prominent Globe icon, eliminating redundant duplicate switches from the model popover.
• Cognitive Controls & Personas: Deep parameters (Creativity / Temperature corridor, Context Buffer depth, Target Personas, and Background models) are managed centrally in the System Config Sheet (Kontext, Modell-Verhalten & Speicher), keeping the input cockpit calm and clutter-free.
• Send Button (▶): Initiates prompt generation.

2. The Briefing Form

The Briefing Form replaces lazy, single-sentence prompts with rigorous constraint design.

The Briefing Form forces the user to deconstruct their strategic goal into specific, isolated dimensions:

• Context (Background): Why are we executing this strategy? What is the current market reality?
• Goal (Objective): What exact metric or deliverable must this process achieve?
• Constraints (Boundaries): What resources are unavailable? What paths are prohibited?
• Audience (Target): Who intercepts the ultimate output?

By capturing intent through these constrained parameters, the engine automatically formats the payloads with extreme precision, driving a massive improvement in the final mathematical alignment of the result before an AI model ever touches the data.

3. File Context Tray

While you can type directly into the chat or Briefing Form, complex problems usually involve existing documents. The File Context Tray (the attachment icon in the input area) allows you to feed external data directly into the Brainstorm Reactor's working memory.

Supported Formats

You can upload multiple files simultaneously. The Brainstorm Reactor supports:

• Documents: PDF, DOCX, TXT, CSV, Markdown
• Images: JPG, PNG, WebP (processed via Vision models)
• Audio: MP3, WAV (uploaded files are transcribed automatically). Additionally, the Brainstorm Reactor supports live audio transcription directly from the microphone — with optional speaker diarization and parallel AI analysis.

How the Brainstorm Reactor Reads Files

When you upload a document, the Brainstorm Reactor does not just search for keywords. It extracts the full text and holds it in context for the active prompt.

This is highly effective for:

• Uploading a 20-page market report and applying a SWOT method to it.
• Attaching a competitor's website screenshot and asking the AI to reverse-engineer their pricing strategy.
• Dropping a CSV of customer feedback and using a sentiment analysis method.

Image Inspection & Desktop Hover Metadata

Attached files and images feature ambient metadata inspection:

• In-Card File Size: Displays the formatted file size (e.g. 308 KB) directly within the resting card next to the format pill.
• Desktop Hover Tooltip: Hovering over an attached image card on desktop surfaces an instant floating spec pill showing the detected aspect ratio (e.g. 16:9 Landscape, 1:1 Square, 9:16 Story / Reel), exact pixel dimensions (e.g. 2048×1189 px), file size, and readiness status.
• Touch & Mobile Ergonomics: On mobile devices and touchscreens, floating hover tooltips remain fully suppressed to maintain a compact, clutter-free resting layout.
• Image Preview Popover: Clicking the preview icon (Eye) opens an in-situ modal displaying the full image draft alongside format and resolution diagnostics.

Smart Paste, Text Transfer & Context De-Duplication

When pasting large text blocks (> 2,500 characters), the system automatically packages the text into a neutral attachment card (Notiz.txt in DE / Note.txt in EN) to prevent textarea scrolling fatigue and keep the input interface clean.

• Unpack to Input Field ([ ↓ ] / [TXT]): To edit the text directly within the prompt textarea, clicking the [ ↓ ] button on the attachment card, tapping the [TXT] pill, or using [ In Textfeld ] in the preview popover transfers the entire file content back into the prompt canvas. The attachment card is atomically removed upon transfer.
• Strict Context Isolation (Zero Token Duplication): The Brainstorm Reactor enforces a strict single-location model: text resides either in the attachment tray or in the prompt textarea, never duplicated in prompt context. Triggering proactive assistants (Recipe Recommendations or Method Match) samples attached documents for matching without leaking document text into the textarea or duplicating tokens.

Context Limits

The amount of text the Brainstorm Reactor can hold in memory depends on the active Text Model (configured in the System Config). High-tier models feature massive context windows capable of processing entire books at once. If you upload more data than the active model can handle, the system will warn you to switch to a model with a larger context window.

Privacy & Persistence

Files uploaded to the Brainstorm Reactor are processed ephemerally for the duration of the session or workflow step. The content is fed into the AI to generate the response, but the raw files are not permanently hosted or exposed publicly.

4. In-Line Slash Commands (/) & Verb-First Staging

While ⌘K acts as the global platform teleporter, in-line slash commands (/) serve as in-situ cognitive verb modifiers directly within the prompt flow.

To eliminate context-switching between typing instructions and selecting cognitive frameworks, typing / at the start of a prompt or on an empty line opens the In-Line Slash Command Popover.

Functional Zones & Catalog Structure

The slash command palette is strictly categorized into four operational zones:

Non-Destructive Staging (Stage before Launch)

Unlike legacy command prompts that execute immediately upon selection, the Brainstorm Reactor enforces Non-Destructive Staging:

1. Selecting an option (via ↵ Enter or Tab) anchors the chosen tool or workflow as an active badge in Zone 1 (Active Context Anchor).
2. The /command trigger token is cleanly removed from the prompt field.
3. Any trailing instructions you typed (e.g. /10x Packaging ideas for sustainable coffee capsules) remain intact in the prompt canvas, allowing you to review and enrich your prompt before dispatching via Enter.

Dual-Zone Collision Resolution (Methods vs. Workflows)

Several cognitive frameworks exist as both single-step direct methods and comprehensive multi-step workflows (for example, Pre-Mortem direct analysis vs. Pre-Mortem Risk Audit 4-step workflow).

The slash popover resolves this ambiguity by isolating items into distinct zones:

• Direct Tools are prioritized in Zone 1 with high-contrast [1-Step] blue badges.
• Workflows appear in Zone 2 with [Workflow] purple badges and an explicit subtitle notice (wechselt auf Workflow-Modus).

This dual-tier display eliminates accidental mode jumps and ensures clear epistemic expectations before prompt submission.

Keyboard Navigation & Shelf Dominance

• Arrow Navigation: ↑ and ↓ navigate cleanly across group boundaries with automatic viewport scrolling.
• Instant Dismissal: Pressing Esc closes the popover without altering the typed prompt.
• Shelf Dominance: Activating the slash popover automatically supersedes the ambient Smart Assistant Shelf (Zone 2) to prevent competing recommendations or layout jitter.