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

What is InterroGate?

InterroGate is a self-hosted application that connects to Anthropic's Claude models 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 answers questions, voices your characters, renders scenes, and checks consistency, all from inside your established canon.

It began life around a single principle: the writer is the creative authority. That principle still governs everything. The core engine never invents canon or makes story decisions for you — it reasons within the world you defined. A newer, optional generative tier can draft prose on request, but it does so grounded in your canon, lands its output as drafts you own and shape, and never commits anything to your world without you.

Think of it less as an AI that writes for you and more as a collaborator who has memorised your entire series bible — and can run scenarios, speak in your characters' voices, catch contradictions, and, where you ask, produce raw material you then make your own.

What it does

The foundation is a consistency engine — a set of ways to know and pressure-test your world. None of these generate manuscript prose; they reason, voice, organise, and verify.

Beyond browsing, the Archive is a set of visual world surfaces, switchable in one click: the Continuity Map (your world as a node graph around any character — relationships coloured by stance, with rings for what they know, suspect, and are blind to, and one click to interrogate or simulate from any node), the Event Map (your timeline as a cause-and-effect lattice in narrative order, with story arcs you author and causal links the AI can propose for review), and the World Atlas (your own map images with pinnable locations). And the Prose Library is the vault for what you've actually written — including a drag-and-drop story board view of acts, chapters, and scenes.

The generative tier

Beyond the consistency engine sits an optional, premium layer that does generate prose — Create. It is a deliberate product feature, and it is built differently from a generic AI writing tool. The differentiator is not raw fluency; it is world-consistency.

Create can run a single interior scene (an autonomous director drives the point-of-view character to the scene's turn) or a genuine multi-party scene (Ensemble — each participant is an independent, grounded character agent, with a casting director choosing who speaks and the Beat Cop checking every turn). An Author Voice profile, analysed from your own Library prose, carries your sound into the output.

What it keeps you in control of

Whether you use only the consistency engine or also the generative tier, the same guarantees hold. These are design choices, not gaps.

• ✓The AI never writes your canon. Nothing — no Simulate insight, no Interrogate line, no Create draft — enters your world's permanent record without you approving it through the canon pipeline.
• ✓Generation is grounded and reviewable. Create only writes from the world you defined, lands its output as drafts, and runs every draft past a canon validator. You decide what survives the cut.
• ✓The core engine doesn't generate prose. Simulate, Interrogate, Observe, and World Builder are analytical and exploratory by design — they help you understand and check, not draft. World Builder takes only what you type.
• ✓You remain the author. Every creative decision — what happens, what it means, what to keep — is yours. The engine is a powerful tool, not a replacement for your judgment.

The context model

Every time you interact with InterroGate, the AI works from layered knowledge. Understanding the layers makes the system predictable.

In practice you build canon over time, think through additions in the brainstorm tier, and set scene context when you want the AI to reason from a specific moment. The system merges the layers — you just use them.

The canon pipeline

The canon pipeline is how something moves from the AI said something interesting to this is now established in my world. AI output is never automatically canon.

1. 01The AI produces output in a session. You notice a detail, an implication, a response worth keeping.
2. 02You flag it. One click sends it to Canon Review. It has no effect on the world yet.
3. 03You review. Canon Review shows conflicts as an honest field-by-field diff — green for new facts, red where it would actually change existing canon, grey for no-ops — and lets you apply only the fields you tick, for every entry type. A partial extraction can never blank what you already have. Lists of facts reconcile per fact: new knowledge adds alongside old instead of replacing it, and a genuine contradiction offers "keep both" as a dated belief update.
4. 04You approve or dismiss. Approving writes the structured entry straight into your world bible or character files. Dismissing removes it with no effect.
5. 05The consequences are surfaced. After a meaningful approval, a "Beat Cop" pass works out what else it implies — who now knows what, which relationships shifted — and proposes those as patches for you to accept or reject, one by one.
6. 06Every decision is reversible. Approvals log the full before-state, so any past decision can be reopened from the Approval Log — the old canon is restored and the item comes back for a fresh call.

Quill — talk to your world

Quill is a conversational surface over your entire canon, reachable from anywhere in the app. It has three capabilities, and the split matters:

The same review discipline applies as everywhere else: previews first, snapshots before writes, and nothing becomes permanent canon without passing through your hands.

Quill has a flip side: the Stage Manager, an out-of-character helper in the same window that answers questions about the app itself — how a feature works, where a button lives — grounded only on the documentation, never your canon. Ask Quill about your world; flip to the Stage Manager when you're asking about InterroGate.

Workspaces & universes

InterroGate keeps multiple isolated world contexts called workspaces — each with its own characters, world bible, prose library, sessions, and canon. Switching in the sidebar reloads the world instantly, with no bleed between projects. Create a new one from the Settings panel; bootstrap one from a public-domain text with the included Gutenberg import tool.

A universe is a shared canon tier that several worlds can inherit — useful for a series with multiple casts, eras, or subseries that share a backdrop. Attach a world to a universe and it inherits that shared canon automatically, in every mode, with no extra cost. Because attaching is a single reversible setting, the same world can be pointed at one universe or another to explore alternate macro-truths — the same scene under two different "what's true at the top level" — without editing a word of the world's own canon.

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 been written. Nothing in it reaches the AI's context unless you choose.

Add entries by pasting, uploading a .txt or .docx (chapters are detected automatically and shown before import — including Act.Chapter.Scene-numbered manuscripts), or saving a session transcript. From the Library you can run an AI scene breakdown, extract world data from your prose into characters and a world bible, run a minimal-touch Polish copy-edit (which never touches deliberate style), and compile selected scenes into a downloadable manuscript — plain text, Markdown, or Word. It is per-workspace, fully searchable, and survives every reset — "Full Wipe" does not touch it. A Board view lays the same manuscript out as a wall of act bands, chapter columns, and scene cards — drag to rearrange, reserve empty slots as stubs, and flip a time lens to see book order versus story order (flashbacks flagged). Reference material (imported world notes rather than manuscript prose) files itself into a collapsed appendix below your chapters.

A growing family of prose passes reads your Library and reports back without ever editing your words: Scene Dating places each scene on your story timeline (flashbacks dated by the events depicted, not the narration), and the Pronoun Consistency lint flags places where the prose contradicts a character's established pronouns — report-only, your prose stays byte-identical. When you do want a change, ReScribe rewrites a single paragraph you select, to your instruction, in your voice, as a reviewable proposal. And Braindump is the bulk flipside of brainstorming — paste any unstructured material (notes, plans, musings) and it is extracted into reviewable world entries.

The Archive has matching hygiene tools for the world itself: a semantic duplicate scan across every category (it catches "the same event under two different names", not just similar spellings) — and, better, duplicates are now stopped at birth: extraction is shown your existing roster, spelling and title variants are matched to the entry you already have, and near-misses arrive as reviewable conflicts instead of forked files. Add to that a reviewable character merge that combines two records of the same person field-by-field and repoints every reference, batch knowledge dating across the whole cast, first-class gender & pronoun fields the renderer actually respects, and relationship stances — a first-class ally→enemy disposition on any relationship, backfillable in one reviewed pass, that colours the Continuity Map and grounds how characters treat each other in scenes.

Accounts & keys

InterroGate runs either as a single-user local app or as a small invite-gated multi-user instance (the current beta). In multi-user mode you sign in, your world data is isolated to your account, and a modest daily budget covers house-key usage (with a larger once-off allowance on your first working day, when world extraction runs). Adding your own key (BYOK) lifts the budget — you spend on your own key, and it is never shared with or visible to anyone else. The key system is open-provider: an Anthropic key is the first-class path, and OpenAI or a local model (Ollama, LM Studio) can be plugged into the same slot. The advanced generative features (Create) unlock via a simple in-app request once your world engine is up and running. You can change your password from Settings at any time (doing so signs out every other device). The application is self-hosted: your world data lives on your machine or your server, never in a third-party store — and it is self-downloadable and self-deleteable at any time.

Two visual themes ship, each invertible dark/light: Spectrum (an accent-driven palette where each module owns a colour from a violet→red spectrum) and Binary (high-contrast greyscale, with only genuine status lights keeping their colour). A live theme editor lets you tune it further.

Working with other tools

Your world is not locked in. Any workspace exports as a World Pack — a portable, tool-neutral file holding your world bible, approved facts, and character sheets — so another writing or world-building tool can import your canon without sharing a line of code with InterroGate.

It works in the other direction too: the Machine Worlds API lets an external story-generation tool upload a World Pack, ask whether the world is complete enough to write from, and submit draft prose for a Beat Cop canon audit — the same validator that guards Create, returning a pass/revise verdict with the exact quotes that break canon and why. Machine worlds live in their own sealed namespace, spend their own key, and never touch your account or budget. The first tool built against it is a companion story studio; the contract is open to others.

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.

The philosophy

InterroGate is built from a specific conviction about what AI should do in a creative workflow — and that conviction survives the addition of a generative tier intact.

There is a meaningful difference between AI as a consistency engine and AI as an unaccountable creative agent. The first is useful to writers. The second undermines them — replacing the writer's voice, producing plausible-but-hollow content, and quietly shifting authorial responsibility away from the human.

InterroGate's answer is to make generation accountable to your world. The consistency engine never invents. The generative tier, when you choose to use it, only writes from your canon, is checked against it, and lands as a draft you own — not a finished page slipped into your manuscript. The moat is world-fidelity, not fluency.

This is why the World Builder has no AI generation, why the canon pipeline requires your approval at every step, why Create's output is always a draft and never canon, and why a validator stands between generation and your eyes. The friction is intentional. The author stays the author.

Getting started

New accounts begin on a blank workspace with two paths: start a world from scratch, or import prose you've already written.

Path A — Start a new story

1. 01Create a workspace. Settings → New Story names a fresh, empty world and switches you into it.
2. 02Add your world. Type characters and world elements in the World Builder, or paste prose into the Library and run Extract World to build characters and a world bible from it. Refine in the Archive.
3. 03Start interrogating. Pick a character in Interrogate mode and ask them something. Response quality tracks the quality of your world data — start simple and refine.

Path B — Import your own prose

1. 01Upload your manuscript. Library → Import accepts .txt and .docx, detects chapters, and lets you merge / split / remove before committing.
2. 02Extract the world. Run the phased extraction over your prose to populate characters, timeline, locations, and concepts — then review the results in Canon Review.
3. 03Explore. Interrogate your cast, simulate "what if" branches, and map what each character knows. When you want raw scene material, Create writes it grounded in everything you just built.

Tips for better results

• Write your character files as if the AI will read them — because it will. Specific psychology, speech patterns, and knowledge limits produce far more consistent voices than vague summaries.
• Set scene context (timeline position + scene) before a moment-specific session, so the AI reasons from exactly the context you're in — and so knowledge filtering withholds anything a character hasn't learned yet.
• Date your characters' knowledge. Marking when a fact was learned lets InterroGate keep earlier scenes honest — a character can't recall something they don't yet know.
• Use Canon Review actively. Capture the precise individual claims you want to hold true, not whole AI responses.
• Export regularly. The ZIP backup is your safety net — there is no undo for resets.

Limitations

InterroGate is in active development (invite-gated multi-user beta). Current practical limits: