Deep Analysis
Some pages deserve more than a quick scan. You find an article that’s directly relevant to your book — a historian explaining how a specific political structure worked, a scientist describing a phenomenon you want to use, a craft essay on a technique you’re trying to learn — and you need to actually understand it, not just save it. The Analysis tab exists for that work. It runs a structured analysis on whatever content is currently loaded in the browser (or on a specific page you point it at), producing a summary, key points, entity mentions, and project-relevance scoring that distill the page into something you can actually absorb in a minute instead of re-reading the whole thing twice. It’s also where Ishvana’s perception engine — a deeper, multi-depth content analysis — lives for the moments when summary-level analysis isn’t enough.
The tab sits alongside Search and Pipeline in the Research panel, and it’s for analyzing a single specific thing rather than searching across your whole research library. If you have a page in the browser, this is the fastest way to get actionable output from it.
Basic analysis
Section titled “Basic analysis”Click Analyze this page on the toolbar and Lagan runs her standard content analysis on the browser’s currently-loaded page. The output is a structured card in the Analysis tab with:
- Summary. 2-3 sentences capturing what the page is about.
- Key points. A bulleted list of the most important takeaways, usually 4-8 bullets.
- Entities. People, places, concepts, events, works cited, technical terms — everything worth flagging.
- Project relevance. A 0-100% score indicating how much the page’s content matches your current project. High relevance suggests the page is directly useful; low relevance suggests it’s tangential.
- Estimated reading time. Based on word count, how long a human would take to read the full page.
- Suggested tags. Auto-generated tags that would be applied if you bookmarked the page.
The analysis runs in seconds and the results can be copied to clipboard in one click. The most common workflow: load a page, analyze it, copy the key points into your outline or notes, close the page, move on.
Deep Perceive
Section titled “Deep Perceive”The Deep Perceive action is a step up from basic analysis. It runs Ishvana’s perception engine on the page, which is a multi-depth content analysis that produces richer output than the basic summary. The output includes everything from basic analysis plus:
- Entity confidence scores. Each entity is flagged with how confident Lagan is that it’s actually a meaningful mention vs. a stray word.
- Relationships. Detected connections between entities. “A mentioned in relation to B,” “A is a type of C,” etc.
- Sentiment analysis. Per-section tone and emotional arc across the page.
- Insights. Lagan’s own observations about what’s interesting or unusual about the content. Not a summary — a commentary on the material.
- Narrative shape. If the page has a narrative structure, the detected arc or argument flow.
Deep Perceive is more expensive than basic analysis (more LLM tokens, longer run time) but the output is meaningfully richer. Use it for content that’s worth the extra cost — long-form articles, dense essays, primary sources you want to really understand.
Perception depth
Section titled “Perception depth”Deep Perceive has four depth levels:
- Surface. Fast perception — essentially a richer basic analysis with entity confidence and relationships added. Takes a few seconds.
- Analytical. Standard perception depth — adds sentiment analysis and insights. Takes 10-20 seconds for a typical page.
- Strategic. Deeper perception — adds narrative shape analysis and explicit argument tracking. Takes 20-40 seconds and uses more tokens.
- Prophetic. The deepest level — adds cross-referencing against your existing knowledge base and speculation about how the content might be relevant to your project in ways you hadn’t thought of. Takes 30-60 seconds and uses the most tokens.
Each depth has a cost hint in the UI — a rough token estimate so you know what you’re committing to before running. The defaults are tuned so Surface and Analytical are cheap enough for routine use and Strategic/Prophetic are reserved for high-value work.
Results display
Section titled “Results display”Results appear as expandable cards in the Analysis tab. Each section of the analysis (summary, key points, entities, insights, etc.) is its own collapsible block. The most important sections (summary, key points) are expanded by default; the deeper sections collapse so you can focus on what matters first.
Each entity in the entities list is a confidence bar — a visual indicator of how confident Lagan is in the detection. High-confidence entities are solid bars; low-confidence entities are faded. You can filter to only high-confidence entities if you want to ignore Lagan’s guesses.
Each insight in the insights list shows Lagan’s own note plus the reasoning she used to generate it. Clicking through an insight expands to the full reasoning. Insights are the most subjective part of the analysis and the one most likely to produce “huh, interesting” moments that genuinely change how you think about the content.
Analysis history
Section titled “Analysis history”Every analysis you run is saved in the project’s research database. The Analysis tab has a sidebar showing your recent analyses, sorted by timestamp. Click any historical analysis to reload its results without re-running.
Saved analyses persist across sessions. If you analyze a page today and come back to it next week, the analysis is still there — and if the page’s content has changed, you can re-run to get a new analysis while keeping the old one for comparison.
When to use Analysis vs. Pipeline
Section titled “When to use Analysis vs. Pipeline”The Analysis tab and the Research Pipeline both process content with an agent, but they do different things.
Analysis works on a single page or piece of content. It’s focused, fast, and produces structured output about that one thing.
Pipeline works on a query that might match content across your whole research library. It searches, consults specialists, optionally runs fresh web research, and synthesizes a context summary.
Use Analysis when you have a specific page in hand and want to understand it. Use Pipeline when you have a question and want to know where the answer lives.
What Analysis is not
Section titled “What Analysis is not”- Not a full-text extraction. The Extract tab handles raw text extraction. Analysis produces structured summary, not raw content.
- Not a translation service. If the page is in a language Lagan can handle, analysis works. For genuine translation of non-English content, the output is approximate and you should use a dedicated translation tool.
- Not fact-checking. Analysis summarizes what the page says. It doesn’t tell you whether the page is correct. For fact-checking, send the claim to the WorldKnowledge agent.
- Not continuous monitoring. Analysis runs on demand. If you want ongoing monitoring of a page for changes, use web monitors in the Research Pipeline tab.