Smart Bookmarks
Regular bookmarks are a list of URLs. You save a page, you label it, maybe you organize it into folders, and six months later you can’t find anything because “interesting article about tides” isn’t specific enough to search for, folders decay faster than you can maintain them, and the original page has probably changed or disappeared anyway. Smart Bookmarks solve this by turning every saved bookmark into structured data — the page’s content is extracted, an agent analyzes it for key points and entities, auto-generates relevant tags, computes reading time, and stores everything in ChromaDB as a vector embedding so future searches can find it by meaning instead of by remembered keywords. The result is a research library that gets smarter the more you use it, instead of a junk drawer that gets worse.
Smart Bookmarks are accessed from the Research panel’s Bookmarks tab, and they’re separate from the ordinary browser bookmarks most apps implement. The key difference is everything that happens after you save — which is a lot.
What happens when you save a bookmark
Section titled “What happens when you save a bookmark”Click the bookmark button on the toolbar while viewing any page, and Ishvana runs a specific sequence of operations in the background:
- Content extraction. The browser’s extracted content (from the cascade described in Embedded Browser) is captured.
- Analysis. The content is sent to Lagan’s analysis pipeline, which runs an LLM prompt with your project’s context (active project name, recent lore keywords, relevant Legendry mentions pre-searched for the page’s content). The prompt asks for:
- A 2-3 sentence summary.
- A bulleted list of key points.
- Entity mentions (people, places, concepts, events).
- Suggested tags based on the content.
- Tag generation. The suggested tags from the analysis are combined with any project-lore keywords that matched the content, producing an auto-tag set.
- Reading time calculation. The word count of the extracted content is divided by 200 (standard reading speed) to produce a reading time estimate.
- Project relevance scoring. The content is scored against your current project — how much it matches your active lore keywords, how many of your characters or places are mentioned, how closely it aligns with topics in your outline. The score is a 0-100% relevance indicator.
- Vector embedding. The full extracted content is embedded by the local embedding model and stored in ChromaDB.
- Metadata persistence. The bookmark’s URL, title, extracted content, summary, key points, entities, tags, reading time, relevance score, and timestamp all get saved to SQLite as a bookmark record.
The whole process takes a few seconds — usually under 10 — and happens in the background. You see a “bookmarking…” toast, then a confirmation that the bookmark was saved, and you can immediately keep browsing. The analysis runs asynchronously; if it’s slow, the bookmark is still saved with the URL and extracted content, and the analysis data gets added to the record when it finishes.
What the bookmark record contains
Section titled “What the bookmark record contains”After a successful save, a smart bookmark has:
| Field | Source | Purpose |
|---|---|---|
| URL | The page | Where to return to |
| Title | The page’s <title> | Human-readable name |
| Domain | Derived from URL | For grouping and filtering |
| Extracted content | Browser extraction | The actual article text |
| Summary | Lagan analysis | 2-3 sentence overview |
| Key points | Lagan analysis | Bulleted list of takeaways |
| Entities | Lagan analysis | People, places, concepts mentioned |
| Tags | Auto-generated + user-added | Searchable labels |
| Reading time | Word count calculation | How long the page takes to read |
| Relevance score | Project match | How useful for current project |
| Created at | Timestamp | When the bookmark was saved |
| Project ID | Active project | Which project it belongs to |
| Vector embedding | ChromaDB | Enables semantic search |
The user-editable fields are title, tags, and an optional notes field. Everything else is generated from the page’s content and isn’t meant to be edited — if the summary is wrong, you can re-run the analysis, but you wouldn’t type a different summary manually.
Semantic search
Section titled “Semantic search”The whole reason the vector embedding matters: semantic search. Ishvana’s bookmark search isn’t keyword-matching the title or tags — it’s matching the meaning of your query against the meaning of the bookmark’s content.
You type “medieval metallurgy techniques” and the search returns bookmarks about:
- Medieval sword forging (obvious match).
- “Damascus steel: the lost art” (semantically adjacent, no keyword overlap).
- “How European swordsmiths learned from the Middle East” (adjacent, different wording).
- “The history of ironwork in the 13th century” (adjacent, different phrasing).
All four would show up in a semantic search for your query. Only the first would show up in a plain keyword search. The semantic search is what makes large bookmark libraries actually useful — when you have 500 bookmarks and you remember “something about medieval craftsmen,” keyword search fails and semantic search finds the thing.
For more on how semantic search works under the hood, see the Semantic Search engine page.
Filtering and organizing
Section titled “Filtering and organizing”Beyond semantic search, the Bookmarks panel supports classic filtering:
- Filter by tag. Type or click a tag to narrow the list to bookmarks with that tag.
- Filter by project. View bookmarks from any project or only the active one.
- Filter by date range. Last 7 days, last month, custom range.
- Filter by reading time. “Short reads under 5 minutes” or “long-form 15+ minutes.”
- Sort by date, project, or alphabetical.
Filters compose. You can look at “all bookmarks tagged ‘history’ in the current project, created in the last 30 days, sorted by date” as a single filtered view.
The bookmarks dashboard
Section titled “The bookmarks dashboard”The Research panel’s start page is the Bookmarks Dashboard — a searchable card grid of all your bookmarks with title, domain, summary, up to 3 auto-generated tags, and creation date visible on each card. Click any card to navigate to that URL in the browser.
Below the dashboard grid, a “Recent” section shows the last 10 pages you visited from the in-session browsing history, so you can quickly return to pages you almost-but-didn’t save.
The dashboard is what you see when you first open the Research tab without a page loaded. It’s the default view.
Bookmark details modal
Section titled “Bookmark details modal”Click the info icon on any bookmark card (or select the bookmark from the Bookmarks tab list) and a detail modal opens with:
- Full analysis — summary, key points, entities, relevance score.
- Full tag list — auto-generated and user-added tags, editable.
- Reading time and word count.
- Actions — open in browser, copy URL, copy content, delete bookmark, add to Lore (creates a reference lore entry), edit tags.
The “Add to Lore” action is one of the most useful. It creates a new Legendry entry in the Reference category with the bookmark’s summary, source URL, and extracted content — turning a research bookmark into a citable lore entry in one click. Useful when a research finding should be part of your canonical project data, not just research reference material.
What happens if the page changes
Section titled “What happens if the page changes”Smart bookmarks store the page content at the time you saved it. If the page later changes (or disappears), the bookmark still has your extracted copy. You don’t lose access to the original just because the web forgot about it.
The URL is still saved, so you can always try to re-visit the page. But if the page is gone, the stored content is your local backup.
What smart bookmarks are not
Section titled “What smart bookmarks are not”A few honest limits:
- Not a full web archive. Smart bookmarks store the extracted text content, not the full HTML, images, or rich media. For a proper web archive, use a dedicated tool like the Wayback Machine or Archive.org.
- Not a replacement for your main browser’s bookmarks. If you want to bookmark a site to open it in Chrome, smart bookmarks don’t sync to Chrome. They live in Ishvana.
- Not cloud-synced. Smart bookmarks are local to your machine. If you want them on another device, you’d need to copy the data directory.
- Not automatic. You have to click the bookmark button. Ishvana doesn’t automatically save every page you visit because that would produce a junk drawer, which is the thing smart bookmarks are designed to avoid.