Lagan
Lagan is Ishvana’s specialist for everything that happens outside your own Legendry. Web research, browser automation, API integration, content extraction, pattern analysis, knowledge processing, YouTube transcription — any time the information you need isn’t already in your project, Lagan is the agent that goes and gets it. Where Hawken coaches your prose and Ishvana orchestrates and Lorekeeper guards your canon, Lagan is the one you send out into the world to bring things back. The research workflow most authors have is this: open a browser tab, read a Wikipedia article, copy a paragraph into a notes file, open another tab, read a forum post, copy another paragraph, open a PDF, bookmark it, forget where you saved it. Lagan replaces that workflow with a single agent that remembers everything it finds, structures it against your project, and hands you the distilled result.
Home module
Section titled “Home module”Lagan lives in the Research module. That’s where you’ll use her most — the Research tab has a dedicated “Lagan Chat” workspace where you have freeform conversations with her about whatever topic you’re researching, and the whole Research module’s pipeline (semantic search, web monitoring, content analysis) is built around her capabilities. When you launch a research query from any other module — click “Research this” on a Legendry entry, ask Ishvana “can you look up X for me” — Lagan is the agent that actually runs the work.
She’s also reachable directly from the Chat tab as one of the selectable agents, and from Writers’ Room mode where multiple agents respond to the same message in parallel. But her home is Research.
What she actually does
Section titled “What she actually does”Five distinct capabilities, each of which exists because a real author workflow needed it.
Knowledge base search
Section titled “Knowledge base search”The first thing Lagan does, before she goes anywhere outside your project, is search what you already have. She runs semantic queries against your project’s ChromaDB vector store — bookmarks, previous research results, extracted content from pages you’ve analyzed before, YouTube transcripts, anything else that’s been indexed. If the answer to your question already lives somewhere in your Research history, Lagan finds it first and cites it before reaching for external sources. This saves tokens, saves time, and avoids the common research-assistant failure mode of re-fetching the same Wikipedia page on every query.
Web research and browser automation
Section titled “Web research and browser automation”When knowledge base search doesn’t find what you need, Lagan goes out. She can run agent-driven content analysis on any page the Research browser is currently displaying — summary extraction, key point identification, entity mention detection, project-relevance scoring. She can add pages as smart bookmarks with auto-tagging. She can read a URL you give her and return a structured analysis without opening the browser at all. And she can set up web monitors that watch a URL for changes and notify you when something new appears.
Every page she analyzes gets stored in ChromaDB with its vector embedding, which means next time you search, that page is part of your knowledge base. Research compounds. You don’t lose things.
API integration
Section titled “API integration”Lagan can hit arbitrary REST APIs with GET or POST requests. Most of this happens through the Research module’s API fetcher, which is exposed in the “Integrate” tab — you paste an endpoint URL, add headers and body if needed, hit fetch, and the response comes back as structured data. Lagan reads the response and can summarize it, extract entities, or cross-reference it against your project.
This is less dramatic than it sounds. It’s not a replacement for a real developer toolkit. It exists for the specific case where a fiction writer needs to look up a specific fact from a specific API — a historical dataset, a language database, a reference API for something niche — without having to learn how to make HTTP requests themselves.
Pattern analysis
Section titled “Pattern analysis”When you give Lagan a document or a piece of text, she can run pattern analysis on it. What are the recurring concepts? What are the entities mentioned? What’s the tone? What’s the structural shape? This is useful when you’re trying to reverse-engineer a piece of inspiration — “what is it about this other author’s opening chapter that makes it work?” — and Lagan can give you a structured breakdown instead of just vibes.
Content extraction and cleanup
Section titled “Content extraction and cleanup”Lagan can take messy source material — a scraped webpage, an OCR’d PDF, a YouTube transcript — and clean it into something you can actually read. Stripping boilerplate, merging fragmented sentences, removing navigation text, fixing paragraph breaks. She’s not an editor; she’s just trying to give you the actual content without the junk.
The Research Pipeline workflow
Section titled “The Research Pipeline workflow”Lagan’s flagship feature is the Research Pipeline in the Research tab. It’s an end-to-end workflow that goes like this:
- You type a research query — “how do medieval city-states actually organize their governments” — and pick which sources you want Lagan to consult.
- Lagan runs semantic search against your project’s knowledge base (existing research, bookmarks, lore entries) and scores relevance.
- If the knowledge base doesn’t have enough, she optionally consults other specialist agents — asking WorldKnowledge for real-world facts, asking Hawken if it’s a writing-craft question.
- She optionally runs fresh web research via the integrated browser and Wikipedia integration.
- Everything gets synthesized into a context summary with a 0-100% relevance score.
- The results are copyable as formatted Markdown, or saveable as a new research bookmark for next time.
The whole pipeline runs with explicit progress streaming — you see what Lagan is doing in each step, which sources she’s consulting, and when she finishes. It’s not a black box. You can cancel mid-stream, inspect intermediate results, and steer her toward specific sources.
YouTube transcription and analysis Research tool
Section titled “YouTube transcription and analysis ”One of the more surprising things Lagan does. Give her a YouTube URL and she can fetch the video’s transcript (if one exists), segment it, run analysis on the segments, and extract the key points, entities, and topics the video covers. The result is searchable like any other research bookmark.
This exists because a lot of research material for fiction writers lives on YouTube now — lectures, documentaries, interviews, world-specific deep dives — and manually transcribing them is miserable. Lagan handles it and the result lives in your knowledge base forever.
How Lagan fits with the other agents
Section titled “How Lagan fits with the other agents”Lagan is one of five divine agents, and her role is specifically complementary to the others:
- Ishvana is the orchestrator. She decides whether a question needs Lagan at all. If the question is about your own world, Ishvana handles it herself. If it’s about something outside your world, Ishvana delegates to Lagan.
- Hawken is the writing coach. If a research finding needs to be turned into prose, Hawken can take Lagan’s output as context and sketch a starting draft for you to rework, or give you craft notes on a draft you’ve already written.
- WorldKnowledge handles real-world factual questions via Wikipedia. Lagan can consult WorldKnowledge for facts she doesn’t want to fetch herself. The division is loose: WorldKnowledge is for clean, trusted factual queries; Lagan is for messy, source-from-anywhere research.
- GameMaster handles mechanics and stat questions. Lagan doesn’t touch mechanics — if you ask her about stat blocks, she’ll tell you that’s GameMaster’s job.
The division of labor is meaningful because each agent has a different system prompt, different tool access, and different output shape. Lagan’s outputs are research summaries with citations; Hawken’s outputs are craft notes and prose suggestions; GameMaster’s outputs are numeric tables. You can’t just ask Hawken to do research — well, you can, but the output will be shaped like writing-craft advice instead of like a research brief.
What Lagan is not
Section titled “What Lagan is not”A few things worth being explicit about.
She doesn’t write your book. Lagan is a research agent. If you ask her “write me a scene where the hero meets the villain,” she’ll tell you that’s outside her domain, point you at Hawken for prose-craft help, and offer to research whatever background the scene needs. The division is intentional — research voice and writing-craft voice are different skills, and forcing one agent to do both produces muddled output.
She doesn’t fact-check your own world. Lagan consults external sources. She doesn’t cross-reference your Legendry for contradictions — that’s what Lorekeeper does. If you want to know whether your prose contradicts your own canon, ask Lorekeeper, not Lagan.
She doesn’t replace real research. Lagan is a research accelerator. She can summarize Wikipedia, she can run analyses on pages you hand her, she can find patterns, she can extract content. What she can’t do is evaluate the credibility of her sources for you — if you’re writing historical fiction and you need to know whether a particular claim is accepted by serious scholars or is a fringe internet theory, Lagan can surface the claim but she can’t tell you which one it is. That’s still on you.