Research Pipeline

The Research Pipeline is the most powerful feature in the Research module, and it’s also the feature most authors don’t realize they need until they use it. Most writing research follows the same frustrating shape — you type a query into Google, you open five tabs, you read four of them, you synthesize the useful bits in your head, you paste some notes somewhere, and you try to remember what you learned a week later when you need it again. The pipeline replaces that whole workflow with a single end-to-end research operation. You type a query once. Lagan searches your local knowledge base first, consults specialist agents if needed, optionally runs fresh web research, synthesizes everything into a structured summary, and hands you back a scored context bundle that you can save as a new research bookmark or copy into your notes. It’s the feature that makes Research feel like a specialized tool instead of just “a browser inside Ishvana.”

The pipeline lives in the Research panel’s Research tab. Most of the features in the rest of the Research module either feed into this pipeline (smart bookmarks become sources) or are side views of what it produces (analysis panels show individual components). If you’re going to do serious research in Ishvana, this is the tab you’ll use most.

How the pipeline runs

You type a query — “how do medieval city-states actually organize their governments?” — and click Run. The pipeline then moves through six steps, streaming progress visibly as it works.

Step 1: Query parsing and term extraction

Lagan’s first job is to turn your natural language query into a set of search terms. She identifies the key nouns and concepts, expands them with related terms where useful, and produces a structured query plan. The plan shows up in the UI as a list of extracted terms so you can see what she’s about to search for. If the terms look wrong, you can cancel and rephrase.

Step 2: Knowledge base search

Before going anywhere external, Lagan searches what you already have. The knowledge base is everything indexed by ChromaDB in your current project — smart bookmarks, lore entries, research notes, cached Wikipedia pages, prior research pipeline results, YouTube transcripts, document content. If your answer is already in your project, the pipeline finds it first.

The search returns ranked matches by similarity, and you see them appear in the UI as they’re found. Each match shows the source (bookmark, lore entry, etc.), the similarity score, and a preview of the matched content.

This step alone is worth the pipeline’s existence. Authors routinely research the same thing twice because they forget they already did it. The knowledge base search catches the duplication and hands you your own earlier research before you waste time re-doing it.

Step 3: Lore context

If you have relevant lore entries in your Legendry — characters, locations, factions, concepts that match the query’s terms — they get pulled in as context. Lore entries are a separate source from bookmarks because they represent your world’s canon, not external research. The distinction matters for the final synthesis: research findings might contradict each other, but lore entries are ground truth for your fiction.

Step 4: Specialist consultation (optional)

If the query is specialized in a way that another agent is better equipped for, Lagan can consult the appropriate specialist:

• WorldKnowledge for real-world factual queries. “How does shotgun gauge actually work” gets passed to WorldKnowledge, who runs a Wikipedia lookup and returns clean factual content.
• Hawken for writing-craft questions. “How do other authors handle multi-POV chapters” gets passed to Hawken, who brings in craft knowledge from his system prompt.
• GameMaster for mechanics questions. “What’s the typical damage range for a crossbow in D&D 5e” gets passed to GameMaster.

Specialist consultation is opt-in via a toggle on the pipeline panel. When on, Lagan decides whether to consult and which specialist to pick. When off, the pipeline runs entirely on Lagan’s own capabilities plus the knowledge base search.

Step 5: Fresh web research (optional)

If the knowledge base and specialist consultation don’t satisfy the query, and you’ve enabled web research, Lagan runs a web search through the integrated browser’s Wikipedia integration or external search. The fresh research gets cached in ChromaDB so the same query next time finds it locally instead of running the search again.

Web research is the most expensive step (actual network calls, LLM token costs) and is off by default. Enable it when you’re doing fresh research and want to go beyond what’s already in your knowledge base.

Step 6: Context synthesis

The final step combines everything — knowledge base matches, lore context, specialist consultation output, fresh research — into a single context summary. The summary is structured:

• Relevance score. 0-100% how much the gathered context matches the query.
• Research terms. The terms Lagan extracted in step 1.
• Knowledge base results. The top matches from your existing research library with similarity scores.
• Lore matches. Relevant Legendry entries with excerpts.
• Specialist findings. Output from any consulted specialists.
• Web findings. Fresh research (if enabled) with source URLs.
• Synthesized summary. Lagan’s own narrative summary integrating everything above.

The synthesis is the thing you take away from the pipeline. It’s not just a list of sources — it’s a structured answer to your query, with the sources visible so you can check them.

Using the results

Once the pipeline finishes, the synthesized context lives in the panel. You have several options for what to do with it:

• Copy as Markdown. One click copies the full context as formatted Markdown to your clipboard. Paste into your editor, your outline, or wherever you need it.
• Save as research bookmark. Creates a smart bookmark with the synthesized context as its content. Next time you run a pipeline query, this saved bookmark shows up in the knowledge base search for related queries.
• Send to Lore. Creates a new Legendry entry in the Reference category with the synthesized context. Useful when the research finding should be part of your canonical project data.
• Save as note. A lightweight save that stores the result as a plain text note in your research notes collection without creating a full bookmark.

The choice of destination matters. Research bookmarks are your searchable research library. Lore entries are canon. Notes are ephemeral. Pick the one that matches how you’re going to use the result later.

Web monitors

The Research tab also hosts the web monitors sub-panel. A web monitor is a URL you want Ishvana to watch for changes. Add a URL, set a check interval (default every few hours), and Lagan’s background service polls the URL periodically. When the content changes, you get a notification and the new version is available for analysis.

Web monitors are for ongoing research where a specific source is important and you want to know when it updates. Tracking a forum thread. Watching a research paper’s errata. Following a news site’s tag page. Monitors save you from manually re-checking pages that might or might not have changed.

The monitors list shows each URL, the last-check timestamp, the next scheduled check, and a status indicator. Pause, resume, or delete any monitor from the list.

Deep Query sub-workflow

Inside the Research tab, there’s also a Deep Query sub-workflow for heavier research tasks. Deep Query is like the main pipeline but with explicit multi-source synthesis — you pick which sources to consult, you pick which specialist agents to involve, and the output is a longer, more thorough report with explicit citations per source.

Use the main pipeline for most queries. Use Deep Query when you need a depth of research that feels like a term paper instead of a quick lookup.

What the pipeline is good for

A few specific use cases:

• Worldbuilding research. You need to understand how medieval cities were organized, how sailing ships were crewed, how a specific disease spread, how a religious order was structured. The pipeline searches your existing notes, pulls in relevant lore, and runs fresh research if you enable it.
• Fact-checking historical details. You wrote a scene set in 13th-century France. You want to know whether your specific details are plausible. The pipeline consults WorldKnowledge and returns grounded facts.
• Cross-referencing against your own work. You’re writing a new chapter and you vaguely remember writing something about the same topic earlier. The knowledge base search finds your earlier scene before you waste time reinventing the idea.
• Comp title research for marketing. You want to know which other authors in your subgenre have written similar books. Feed the query to the pipeline and get a ranked list of comps that you can then verify manually.

What the pipeline isn’t

• It isn’t a replacement for deep research. The pipeline accelerates research; it doesn’t replace the careful reading required for high-stakes topics. If your book’s plausibility depends on getting a specific historical fact exactly right, you still want to read primary sources.
• It isn’t a fact-checker. Pipeline results are synthesized from whatever sources were consulted. If your knowledge base has wrong information, the pipeline can surface it with confidence. Double-check claims that matter.
• It isn’t a citation manager. Pipeline results include sources, but Ishvana doesn’t format citations for academic publication or manage a bibliography.

What’s next

Search The standalone semantic search tab — simpler but faster for quick lookups.

Smart Bookmarks The knowledge base the pipeline searches first.

Lagan Chat Freeform conversation with Lagan when you want dialogue instead of structured pipeline output.