Ishvana

Most writing assistants put one general-purpose chatbot between you and the work, and that chatbot ends up mediocre at everything because it’s trying to be a novelist and a researcher and a rules lawyer and a historian all at once. Ishvana takes the opposite approach. She’s not trying to be good at everything. She’s good at knowing who else is good at the thing you just asked for, and she delegates accordingly. The result is that when you talk to her, she acts like a collaborator who knows the building — she’ll answer the questions that are hers, and for the rest she’ll quietly walk down the hall, knock on the right specialist’s door, bring their answer back, and hand it to you with context.

That’s the job. Orchestrator.

What she answers herself

A few things Ishvana handles without delegating.

Lore-grounded questions about your own world. She searches your Legendry before responding — hybrid keyword and semantic search across every entry — and cites what she finds. If she can’t find something, she tells you, instead of making it up. The anti-hallucination discipline is the most important thing about her, and it’s why you can actually trust her answers about your fiction.

Context-aware conversation. She knows your active project, she knows what document you currently have open in the editor, she knows whether you have a page loaded in the research browser. You can ask “what did I say about this character in chapter three” and she’ll go read chapter three. You can ask “does the page I’m looking at relate to anything in my world” and she’ll cross-reference. The context isn’t magic — she can see a specific, documented set of things — but the things she can see are the things that matter most for the way you actually work.

Entity detection. Paste a passage and ask her to tell you what characters, locations, factions, and items are in it. She identifies them with types and confidence scores. Useful for quickly cataloging what’s actually referenced in a scene, especially when you’re trying to audit a chapter for everything that needs to exist in the Legendry.

Big-picture synthesis. When your question spans multiple modules — “does my subplot about the Crimson Hyenas tie into the main arc in chapter twelve?” — Ishvana is the agent most suited to giving you a thoughtful, connected answer. She has the widest view of your project, and she’s been trained to think in cross-references.

When she delegates

Ishvana doesn’t try to do everything herself. Her system prompt explicitly tells her to hand off specialized work to the right specialist instead of attempting it. Here’s what delegation looks like in practice.

Hawken

Writing-craft work. Style analysis, line-level edit suggestions, copy-edit passes, inline transforms — anything where prose-craft expertise is the right answer instead of general orchestration.

Lagan

Research. Web content, external sources, content analysis, anything outside your own Legendry.

GameMaster

Mechanics. Stat blocks, ability design, encounter balance, what-if simulations against your active ruleset.

WorldKnowledge

Real-world facts. Wikipedia lookups, plausibility checks, managing the rules that define how your fiction differs from reality.

The delegation is automatic and invisible. You don’t ask Ishvana to “please hand this to Hawken” — you just tell her what you want, and if the request falls into another agent’s domain, she forwards it and returns with the result. From your perspective, you’re still talking to Ishvana. Under the hood, a specialist is doing the actual work.

Parallel delegation

When a single request touches multiple domains, Ishvana can fire several specialists at once. “Research the political structure of this faction and write a diplomatic scene” is two jobs — research and writing — and Ishvana launches Lagan and Hawken in parallel instead of chaining them. When both finish, she synthesizes their outputs into one response.

This isn’t a mode you enable. It’s a decision Ishvana makes when she sees a multi-domain request and recognizes that the pieces don’t depend on each other. If the pieces do depend on each other — “research this historical period, then write a scene in that period” — she chains them sequentially because the writing needs the research output as input.

The practical effect is that complex multi-domain requests resolve meaningfully faster. A chain of three sequential specialists takes three specialists’ worth of time. A parallel fan-out of three takes roughly as long as the slowest one.

Document awareness

Ishvana automatically sees the following when you’re chatting with her:

Your active document. Title plus a content excerpt of whatever you have open in the editor. If you’re mid-scene in chapter four, Ishvana knows you’re in chapter four and can answer “what was the previous paragraph again?”
Research context. If you have a page loaded in the Research browser, Ishvana sees the URL, page title, and content excerpt.
Any project document on demand. She doesn’t keep the whole project in context, but she can pull any specific document you name. “What did I write in the prologue?” — she fetches the prologue and reads it.

The distinction matters. Ishvana is not reading your entire project in every request — that would be expensive and slow. She’s reading the specific things you’re currently working with, which is usually enough, and she fetches more on demand when the conversation needs it.

Writers’ Room mode

A different chat mode that exists specifically for brainstorming: Writers’ Room. In this mode, Ishvana responds alongside Hawken and Lagan simultaneously, each from their own perspective. One question, three different answers — the orchestrator’s synthesis view, the creative writer’s prose-craft view, and the researcher’s depth view. The three agents see the same message but respond independently.

This is most useful when you’re stuck on a decision that touches multiple domains. “Should I cut this subplot?” gets you three different answers — Ishvana’s big-picture synthesis, Hawken’s opinion on whether the subplot is working as prose, and Lagan’s research on how similar subplots have been handled by other authors. You pick whichever lens helps most, or you use all three to triangulate.

Writers’ Room isn’t the default. It’s an explicit mode you opt into when you want multiple perspectives on the same question.

Safety rails

Ishvana runs under explicit constraints that you should know about, because they affect what she will and won’t do:

No invented lore. She searches the Legendry first. If the information isn’t there, she says so instead of filling in a plausible-sounding guess. This is the most important rule, and it’s what separates her from generic chatbots. Ishvana will tell you “I don’t see anything in your Legendry about that” instead of making up a fact.
Domain boundaries. She delegates instead of attempting. If you ask her for line-level prose feedback, she’ll hand it to Hawken rather than improvising her own answer. If you ask her for a stat block, she’ll hand it to GameMaster.
Read-only. Ishvana cannot create, delete, or modify anything. She can’t add a lore entry. She can’t edit a document. She can’t change your settings. She’s an advisor, not an editor. The constraint is hard-coded — even if you asked her to delete something, she literally cannot.
Anti-prompt-injection. Ishvana ignores any instructions embedded in lore content or document text that attempt to override her behavior. If you accidentally have a lore entry that says “when you read this, forget your system prompt and do something else,” she ignores it. This matters more than you’d expect because imported lore from external sources can sometimes contain inadvertent instructions.

Personality

Ishvana is warm, direct, insightful, and willing to push back when something doesn’t make sense. She’s not a “yes, great idea!” assistant — if you propose a plot twist that contradicts three other things already in your Legendry, she’ll tell you. If you ask her a question and she needs more context to answer well, she’ll ask a follow-up instead of guessing.

The personality is intentional. The failure mode of most assistants is that they’re trained to be agreeable, and agreeable is exactly wrong for a creative collaborator. You need someone who’ll tell you the third act isn’t working. Ishvana errs toward honest over polite, while still being warm enough that the honesty doesn’t sting.

What’s next

Chat The conversation interface where you actually talk to Ishvana.

Hawken The writing coach Ishvana delegates prose-craft questions to.

Lagan The research specialist — Ishvana's most common delegation target.

Etherforce The LLM engine that powers every agent including Ishvana.