Web Search transforms the Brainstorm Reactor from an isolated reasoning engine into an empirically grounded research engine.

Activation

Web search is engaged through two focused interaction points — right where the user needs it:

1. Direct Mode Input Field (SEARCH Mode)

In Direct Mode, click the Search button (globe icon) located directly in the lower-right action bar of the chat input field (ReactorInputArea). When active:

• A blue SEARCH MODE badge appears directly above the input box.
• Every prompt you send includes real-time web research synthesized into the response.
• Click the badge or the globe icon again to deactivate and return to standard reasoning.

2. Autonomous Deep Research Agents

When executing investigative recipes or launching the Deep Research Agent, web search is activated automatically. The agent orchestrates a parallel Web Swarm that queries primary sources, analyzes live documents, and extracts verified citations without manual intervention.

---

How It Works

When Search is active:

1. Query Formulation: The Brainstorm Reactor translates your prompt into targeted, multi-vector search queries.
2. Real-Time Crawling: It fetches and parses relevant web pages, news articles, and primary sources.
3. Synthesis & Grounding: The AI synthesizes the gathered empirical findings with its analytical frameworks.
4. Source Attribution: Exact URLs and domain citations are included at the conclusion of the response.

---

Search Model Architecture

Web Search utilizes a dedicated Search Model (SLOT_SEARCH / fallback_research) — configurable in the System Config sheet under Orchestrator Slots. This operates independently of your primary synthesis model, allowing you to pair a deep reasoning model (e.g. OpenAI o3 / GPT-5.6) with a search champion (e.g. Gemini 3 Flash / Perplexity) for optimal latency and fact extraction.

---

When to Use

• Current market events, news, regulatory updates, and recent developments.
• Competitive intelligence with live corporate data and pricing.
• Fact-checking hypotheses against primary external documentation.
• Researching niche technical topics outside baseline training cutoff dates.

When Not to Use

• Pure creative brainstorming (external web results can induce premature convergence).
• Highly confidential strategic drafting where external search query formation is unnecessary.
• Ultra-low latency tasks where immediate mathematical or logical deduction is the sole priority.