InterroGate — Plain Language Overview

A world-building and narrative simulation engine for fiction writers.

InterroGate is a tool that helps you stress-test your world, interrogate your characters, and maintain consistency across a complex narrative — without writing your story for you.

What is InterroGate?

InterroGate is a local application that connects to Anthropic's Claude AI and uses everything you know about your story world as its operating context. You give it your characters, your timeline, your factions, your locations — and it uses that information to answer questions, voice your characters, and render scenes that are consistent with your established canon.

It is built around a single principle: the writer is the only creative authority. InterroGate does not generate story content, make narrative decisions, or invent canon. What it does is act as an intelligent sounding board that knows your world as well as you do — and can respond from within it.

Think of it less like an AI writing assistant and more like a collaborator who has memorised your entire series bible and is available at any hour to run scenarios, speak in your characters' voices, and check whether a planned scene makes sense given what's already established.

What it does

InterroGate organises its capabilities into four core activities:

◈

SIMULATE

Ask what would happen if a given event occurred. The engine reasons causally from the world state — who would react, how, and why. Useful for pressure-testing plot logic before you commit it to the page.

◉

INTERROGATE

Put a character in the room and question them. The AI speaks in their voice — with their psychology, their knowledge limits, their patterns of speech — and holds a consistent persona across an extended conversation.

◎

OBSERVE

Describe a scene setup and receive a prose rendering in the author's voice. Useful for exploring how a scene might feel — its atmosphere, tempo, and emotional texture — before writing it yourself.

◫ ◌

BUILD & ORGANISE

Add characters, events, locations, and concepts to your world through the World Builder. Browse and search the full canon in the Archive. Capture interesting AI output and move it through a review process before it becomes established fact.

What it doesn't do

This matters as much as what it does. InterroGate is deliberately limited in scope — and the limitations are design choices, not gaps.

✕ It doesn't write your novel. InterroGate is a simulation engine, not a drafting tool. Its output is context, exploration, and consistency-checking. Your prose is yours.
✕ It doesn't generate story content. Simulate, Interrogate, and Observe responses are analytical and exploratory. They are not scenes, not chapters, not copy-paste material. The World Builder explicitly prohibits AI generation — you type your own world data.
✕ It doesn't make creative decisions. The AI does not choose what should happen, who should die, what the story means, or where the narrative should go. It only reasons within the world you have defined. You remain the author.
✕ It doesn't replace the writer. Nothing in InterroGate removes the writer from the process. Every piece of world data in the system was written by you. Every decision about what to accept, reject, or use is yours. The AI is a tool — a sophisticated one, but a tool.

The three-tier model

Every time you interact with InterroGate, the AI is working from three layers of knowledge. Understanding this makes the system much more predictable and useful.

Tier 1 — Permanent

Canon

Everything your world definitively knows. Your character files, your world bible, your approved facts. This is always present. It cannot be overridden by the other tiers.

Tier 2 — Working

Brainstorm

What you're currently developing. Draft ideas, speculative additions, things you've added to the World Builder but haven't approved as canon yet. Available to the AI as working context, clearly not yet settled.

Tier 3 — Immediate

Scene

What's happening right now in the narrative. The timeline position, the scene description, who's in the room. Sets the immediate context for each query.

In practice: you build your canon over time, use the brainstorm tier to think through new additions, and set scene context when you want the AI to reason from a specific narrative moment. The system handles merging these layers — you just use them.

The canon pipeline

The canon pipeline is how ideas move from the AI said something interesting to this is now an established fact in my world.

Generative AI output is never automatically canon. Nothing the AI says becomes part of your world without you explicitly deciding it should. The pipeline enforces this:

01 The AI produces output in a Simulate, Interrogate, or Observe session. You notice something — a detail, an implication, a character response — that feels true and worth keeping.
02 You flag it for review. One click sends that text to the Canon Review panel in the sidebar. It sits there waiting — it has no effect on the world yet.
03 You review and edit. In the Canon Review panel, you can edit the flagged content before approving it. Often the raw AI output needs to be shaped into a clean statement of fact.
04 You approve or dismiss. Approving saves the item to your permanent canon log and includes it in all future context. Dismissing removes it with no effect.
05 You materialise it. Approved items in the log can then be folded back into the structured world bible files and character JSON files at your pace. This step is deliberate — it keeps the canon tidy and under your control.

The key principle

You approve every single fact that enters your world. The AI never writes canon. You write canon — sometimes with the AI as a starting point.

Workspaces

InterroGate supports multiple isolated world contexts called workspaces. Each workspace is a completely separate environment — its own characters, world bible, prose library, and session history. Switching between workspaces in the sidebar reloads the entire world context instantly with no data bleed between worlds.

This lets you maintain multiple projects in parallel (or run a reference world alongside your original work), or use a published fiction world for testing and practice while keeping your original work in a separate, private workspace.

New workspaces can be bootstrapped from Project Gutenberg texts using the command-line import tool — a single command downloads the text, detects the story boundary, and runs three Claude extraction passes to produce a complete set of character files and world bible data. A --prose-only mode does the same without Claude (no API cost) and just sets up the prose library, ready to explore.

Reference workspace

The Silver Blaze workspace (Arthur Conan Doyle, 1892) ships as the reference test environment — 11 characters, 19 timeline events, 11 locations, all extracted automatically from the Gutenberg text.

Prose Library

The Library is a persistent vault for actual written prose — scenes, chapters, fragments. It is deliberately separate from the world bible: the world bible holds intent and canon; the Library holds what has actually been written. Nothing in the Library is injected into the AI's context unless you explicitly choose to do so.

Library entries can be added by pasting text, uploading a .txt or .docx file, or saving a session transcript directly as a Library entry. Chapter detection is automatic — upload a full text file and InterroGate splits it at heading boundaries and shows you the chapters before committing. Gutenberg-formatted texts (double blank lines between paragraphs) are rendered correctly.

The Library is per-workspace, survives all reset operations ("New Story" does not touch it), and is fully searchable across titles and prose content.

Who it's for

InterroGate is built for anyone working with a complex web of facts, people, timelines, and relationships — where internal consistency matters and where being able to test logic before committing to it saves time. The primary use case is fiction, but the underlying capability maps directly to non-fiction domains that share the same structural problem.

Novelists Especially those working in speculative fiction, sci-fi, fantasy, or any genre where the world itself is a character. Useful from early development through late drafting.

Screenwriters Series writers who need to maintain consistency across multiple episodes, seasons, or timelines. Character voice testing is particularly useful for dialogue work.

Game Designers Narrative designers and worldbuilders working on RPGs, video games, or tabletop systems with deep lore. Archive and simulation are natural fits for game development workflows.

Worldbuilders Writers who build worlds as a creative practice in their own right, separate from any specific story. InterroGate gives a world a testable, interactive dimension.

Investigative Journalists Long-form investigations involve dozens of sources, competing timelines, and claims that must hold up across months of reporting. InterroGate's knowledge state model maps directly to source management: what a source knows, what they've disclosed, what they're concealing. The Interrogate mode mirrors actual journalistic technique — you question a source profile built from your notes and test whether their account is internally consistent. When new documents surface, Simulate mode helps you reason through what they imply before you pick up the phone.

True Crime & Narrative Non-Fiction Writers Real events with real people demand the same consistency rigour as fiction — often more, because the record can be checked. When you're reconstructing a timeline from court documents, interviews, and physical evidence, InterroGate's archive keeps the competing accounts straight. Interrogate a person of interest built from what the record says about them. Find where the narratives diverge. The canon pipeline means you're always working from what's established, not what you half-remember from three months ago.

Documentary Filmmakers Documentary development involves tracking multiple subjects across extended shoot schedules, managing interview transcripts against a developing editorial thesis, and ensuring the story you're cutting holds together causally. InterroGate lets you model your subjects, interrogate them from your research, and stress-test the through-line before you're in the edit suite discovering a gap.

Tabletop Game Masters Running a campaign with a living world, factions with agendas, and NPCs who remember what the players did last session. InterroGate gives GMs an interactive world model they can query mid-session or use to plan the next one. Put an NPC in Interrogate mode and discover how they'd actually respond to the players' approach — rather than deciding in the moment and contradicting something you established two sessions ago.

Researchers & Analysts Anyone mapping a complex domain — political movements, corporate structures, historical periods, legal cases — where the goal is to understand how the parts relate and what happens if something changes. Simulate mode is particularly useful for scenario planning: given what you know about these actors and this situation, what are the plausible paths forward? The archive keeps the domain model accessible and queryable as it grows.

Legal & Case Researchers Building a case narrative means managing witness accounts, physical evidence, timelines, and jurisdictional constraints — all of which must be internally consistent and traceable to source. InterroGate's archive and knowledge state model support exactly this: each witness has what they know, what they claim, and what the evidence suggests they couldn't know. Simulation helps identify where accounts are incompatible before they're tested in court.

The philosophy

InterroGate was built from a specific conviction about what AI should and should not do in a creative writing workflow.

There is a meaningful difference between AI as a consistency engine and AI as a creative agent. The first is useful to writers. The second tends to undermine them — replacing the writer's voice, generating content that is plausible but hollow, and slowly shifting authorial responsibility away from the human.

InterroGate takes the first position firmly. It is a pressure-tester, a world-mirror, a character simulator. It knows your world and can reason from within it. What it cannot do — and is deliberately prevented from doing — is make creative decisions or generate content you would put directly into your manuscript.

The design principle

If an InterroGate session ends and you have written more of your own story than you started with, something has gone wrong. A good session ends with you understanding your world better — not with the AI having written it for you.

This is why the World Builder has no AI generation. This is why the canon pipeline requires your approval at every step. This is why Simulate, Interrogate, and Observe outputs are framed as analytical responses, not prose to be lifted. The friction is intentional. The writer's voice remains the writer's.

Getting started

There are two paths: bootstrapping from a public-domain text (the fastest way to see the full system in action), or building your own world from scratch.

Path A — Gutenberg Bootstrap

01 Run the import tool. From the command line: python3 tools/gutenberg_import.py --story "Silver Blaze" --next-story "The Adventure of the Cardboard Box" --output workspaces/silver_blaze. This downloads the text, extracts the story, and runs three Claude passes to produce characters and world bible data. Add --prose-only to skip the Claude extraction (free, no API cost) and just set up the prose library.
02 Switch to the workspace. In the sidebar, click the workspace switcher and select your new workspace. InterroGate reloads with the extracted world context and auto-imports the prose into the Library.
03 Start interrogating. Select a character in Interrogate mode and ask them about the events of the story. Select "YOU ARE" to play a character yourself and conduct a direct interrogation. The extracted world data gives the AI a solid foundation from the first question.

Path B — Original World

01 Create a workspace. Create a folder in workspaces/ with your project name and add a meta.json file: {"display_name": "Your World"}. Switch to it in the sidebar.
02 Add your world bible data. Create JSON files in the workspace's world_bible/ directory covering your timeline, factions, geography, and concepts. Add character files to characters/. The format is flexible — even rough notes formatted as JSON produce useful results. Click "Reload" in the sidebar to pull in changes without restarting.
03 Start interrogating. Switch to Interrogate mode, select a character, and ask them something. The quality of responses is directly proportional to the quality of the world data you've provided. Start simple and refine as you go.

Tips for better results

Write your character files as if the AI will read them — because it will. Clear, specific descriptions of psychology, speech patterns, and knowledge limits produce far more consistent character voices than vague summaries.
Use the Scene Context panel before a session involving a specific narrative moment. Setting timeline position and scene description focuses the AI on exactly the context you're working in, rather than reasoning from the whole timeline at once.
Save sessions when you've reached a useful working state. Session files capture everything — conversation history, scene context, brainstorm items. Coming back to an interesting thread later is much easier if you've saved.
Export regularly. The ZIP backup export is your primary safety net. Run it before any significant world bible changes, before any reset operation, and periodically as a general practice. The files are small and the export is fast.
Use the Canon Review panel actively, not as an afterthought. The best use of the pipeline is capturing specific, precise observations — not whole AI responses, but the individual claims within them that you want to hold as true.

Limitations

InterroGate is a first-generation tool in active development. These are the current practical limitations you should be aware of:

API KEY

InterroGate requires an Anthropic API key. You pay for API usage directly to Anthropic at their standard rates. The application itself is free and open source. You can set the key in Settings; there is no subscription or account required beyond the Anthropic API access.

SELF-HOSTED

InterroGate runs on your own infrastructure — either locally on your computer or self-hosted on a VPS for access from any device. There is no managed cloud service, no account system, and no third-party data storage. Your world data stays under your control. Running it locally requires a Python environment; VPS deployment requires a Linux server and the included setup script.

NO CLOUD SYNC

If you work across multiple machines without a VPS, you will need to manage syncing your world data yourself (e.g., using Dropbox, Syncthing, or git). The ZIP export/import feature is designed to make this practical. A VPS deployment effectively solves this — your data lives on the server, accessible from anywhere.

CONTEXT WINDOW

Very large world bibles — hundreds of thousands of words of context — may exceed the AI's context window or produce degraded results as context density increases. For most fiction projects at any stage of development, this is not a practical concern. If you notice degraded consistency, consider splitting your world bible into more focused files and reviewing what's actually being loaded.

MANUAL CANON

Approved canon items in the log do not automatically update your world bible JSON files. You review the log and make those updates yourself. This is intentional, but it does mean the workflow requires some discipline to keep the canon log and the source files in sync.

NO UNDO

There is no undo system for data operations. Reset and wipe operations are permanent. Export a backup before any significant operation. The export is fast and the files are small — there is no good reason not to.