Research Hub
One screen, two research modes — turn a plain-language prompt into a cited web research report, or search your own Knowledge Base by meaning instead of keywords.
What Is Research Hub?
Research Hub is a single screen with two switchable research modes: Web Research, which scrapes and synthesizes live sources from the web into a cited report, and Semantic Search, which searches your own uploaded Knowledge Base by meaning using AI embeddings. Both modes share the same simple prompt box, so switching between "research the outside world" and "search what we already know" is a single toggle, not a different screen.
The screen is designed for two very different users at once. A non-technical user can type a plain-language question and press go — no configuration required. A power user can open the advanced settings panel to point Web Research at a specific target URL, or tune how deep and how wide the crawl goes.
In one sentence
Type what you want to know — Research Hub either goes out and researches the live web for you, or ranks your own Knowledge Base sources by relevance, depending on which mode you have selected.
Web Research
A backend web scraping/crawling engine finds and reads live pages relevant to your prompt, then an LLM synthesizes what it found into a structured, cited markdown report.
Semantic Search
Your prompt is embedded and compared against your Knowledge Base sources using cosine-similarity, so results are ranked by meaning, not just keyword overlap.
Advanced Settings
Optional panel for power users: a specific target URL, maximum crawl depth, and maximum pages to visit during Web Research.
Save as Report
Web Research results can be saved directly into the shared report system and reviewed later from My Reports.
How It Works
Both modes start from the same prompt box. What happens next depends on which mode is selected:
In Web Research, a backend ingestion engine (a web scraping/crawling engine) discovers and reads pages relevant to the prompt, then an LLM turns the gathered material into a structured markdown report with citations back to the original sources.
In Semantic Search, the prompt is converted into a vector embedding by a local Ollama model, and that vector is compared against every source already stored in your Knowledge Base using cosine-similarity — so the results reflect what the sources actually mean, not just which words they contain.
Web Research Mode
For most users, Web Research needs nothing but a prompt: describe what you want researched, and the engine finds sources on its own, reads them, and writes a cited report. Power users can open the advanced settings panel to take more control:
| Setting | What it does |
|---|---|
| Target URL | Optional. Leave it empty and sources are auto-discovered from the prompt. Set it to point the crawl at one specific site instead of letting it search freely. |
| Max Crawl Depth | How many link-hops away from the starting page(s) the crawler is allowed to follow. |
| Max Pages | A hard cap on the total number of pages the crawl will visit, so a broad prompt can't run away. |
Saving results
Once a Web Research report is generated, use Save as Report to store it permanently. See the "Reports" section below for where it appears and who can see it.
Semantic Search Mode
Semantic Search answers questions against your own uploaded Knowledge Base sources instead of the live web. Your prompt is embedded and ranked against every source's embedding using cosine-similarity, so a question can surface a relevant source even if it doesn't share exact keywords with it.
This mode depends on a local embedding model served through Ollama. If no embedding model is configured for the tenant, Semantic Search gracefully degrades — it still returns results from the Knowledge Base, but they are unranked (returned in their stored order) instead of sorted by relevance. See Embedding Model for setup instructions, and Semantic Search Overview for the underlying concepts (embeddings, vector similarity, and how ranking works in more depth).
No ranking without an embedding model
If results seem to come back in a strange or seemingly random order, check whether an Ollama embedding model is configured — that's the most common cause of Semantic Search falling back to unranked results.
Saving results
Semantic Search results can be saved too — use Save as Report to store the ranked sources as a report. See the "Reports" section below for where it appears and who can see it.
Reports
Both modes support Save as Report. Web Research results are written into TotalApp's shared report system with their source marked as Web Search; Semantic Search results are saved the same way with their source marked as Semantic Search. Either way, the saved report shows up alongside your other saved reports in My Reports.
Every report you save — from Research Hub or any other screen — starts out personal, visible only to you. From My Reports, you can Share a report either with your whole workspace or with specific teammates; once shared, it also appears in their My Reports list.
Personal by default, shareable on demand
Both Web Research and Semantic Search results can be saved as reports. A saved report is visible only to you until you explicitly share it from My Reports — either with the whole workspace or with individually chosen people.