Aevryn

AI-Powered Story Continuity Engine

Built by Aetherra Labs

Primary domain: https://aevryn.ai

Upload a story. Aevryn builds a living canon that stays consistent across every chapter.

Evidence in. Canon out.

Overview

Aevryn is an AI-powered Story Continuity Engine designed for creators who work with novels, manga, manhwa, web novels, comics, scripts, and other long-form stories.

Its purpose is not to generate stories.

Its purpose is to understand existing stories.

Aevryn continuously analyzes chapters, tracks characters, locations, items, relationships, timeline progression, and world state, building a living canon database that evolves alongside the story.

This enables creators to generate consistent scene descriptions, prompt packs, narration prompts, and future AI-assisted media without continuity errors.

Why Not Just ChatGPT?

ChatGPT can answer from a prompt.

Aevryn maintains evidence-backed story state.

Instead of asking an AI to remember everything every time, Aevryn imports story evidence, accepts or rejects extracted candidates, versions canon changes, and reconstructs what was true at a specific chapter or scene.

That is the difference:

Story evidence
-> Candidate facts
-> Canon validation
-> Timeline-aware continuity
-> Human-readable production outputs

Current Status

Aevryn is currently V1 Engine Complete for the terminal-based story continuity workflow.

V1 understands story structure, evidence anchors, candidate extraction payloads, canon updates, timeline-aware scene state, character sheets, world sheets, continuity reports, prompt packs, deterministic exports, and validation runs.

V1 does not include a website, accounts, payments, cloud sync, image generation, video generation, or AI chat.

Trust Signals

Final V1 acceptance validation:

• V1 Engine Complete
• Final V1 acceptance sweep passed
• 903 automated tests passing
• 162 frontend tests passing
• Deterministic rebuild validation
• Cross-genre validation
• 8-genre validation corpus
• 80 local chapter files
• 8,828 evidence anchors
• Evidence-backed canon
• Timeline-aware continuity
• V1 architecture complete
• Validation score: 100%

Final V1 acceptance snapshot:

snapshots/v1_final_acceptance_20260626_122543

Current validation digest:

1a04733548822f82afc473a468cf6a6189387cb97eb915cc803d6b6e94b5d749

---

The Problem

Current AI-assisted recap and storytelling workflows suffer from major continuity problems.

Characters suddenly change appearance.

Weapons disappear.

Buildings revert to earlier states.

Dead characters reappear.

AI has no persistent understanding of the story.

Every prompt starts almost from scratch.

---

The Solution

Aevryn creates a continuously evolving Story State.

Instead of asking AI to remember everything every time, Aevryn maintains the current canonical truth.

Every generated prompt references the latest known state of the story.

---

Core Features

• Story analysis
• Character tracking
• Relationship tracking
• Equipment tracking
• Inventory tracking
• Location tracking
• Building and environment tracking
• Timeline management
• Canon database
• Character card generation
• Scene breakdown
• Canon-backed prompt pack generation
• Narration prompt generation
• Export to external tools

---

V1 CLI Quickstart

Install the project in editable mode:

pip install -e .[dev]

Inspect a text chapter import:

aevryn import path\to\chapter_001.txt --source-id my_story

The import command prints chapter IDs, scene IDs, a scene map, and a bounded evidence-anchor preview. Use scene IDs with --scene-id to inspect timeline-safe character, scene, world, and prompt views.

Generate the evidence-bounded extraction prompt for one scene:

aevryn extraction-prompt path\to\chapter_001.txt --source-id my_story

Apply an AI extraction response from JSON:

aevryn extract-ai-json path\to\chapter_001.txt path\to\ai_response.json --source-id my_story

The command prints accepted and rejected candidate counts plus stable accepted IDs, such as accepted_entity_ids. Use those IDs in follow-up commands like character, scene, world, and prompt.

Generate canon-backed outputs from that response:

aevryn character path\to\chapter_001.txt --source-id my_story --ai-response-file path\to\ai_response.json --character-id character_mark
aevryn scene path\to\chapter_001.txt --source-id my_story --ai-response-file path\to\ai_response.json
aevryn prompt path\to\chapter_001.txt --source-id my_story --ai-response-file path\to\ai_response.json
aevryn continuity path\to\chapter_001.txt --source-id my_story --ai-response-file path\to\ai_response.json

Markdown output is presentation-first for quick scanning. JSON and CSV outputs preserve machine-readable detail for audits, tests, and downstream tools.

Run the validation corpus:

aevryn validate --summary-only

The default validation source root is:

~/Desktop/Aevryn test chapters

Use --source-root or AEVRYN_VALIDATION_ROOT to point validation at a different local corpus.

Create a deterministic validation snapshot:

aevryn validate --summary-only --snapshot-dir snapshots/my_run

Snapshot artifact directories are ignored by git. They store deterministic validation metadata only, not chapter text.

---

Development Checks

Run these before committing:

ruff check pyproject.toml docs src tests validation
mypy src tests
pytest -q
aevryn validate --summary-only

---

Design Philosophy

Aevryn never invents canon.

Every extracted fact should include:

• Evidence
• Source Chapter
• Confidence
• Current Status

If evidence does not exist, the system records the information as Unknown rather than hallucinating details.

---

Long-Term Vision

Aevryn aims to become the industry's most accurate AI-powered story continuity platform.

The long-term goal is to allow creators to upload an entire story and immediately begin producing canonically accurate AI-assisted content with minimal manual work.

---

Built by Aetherra Labs.