Scriptirix AI

AI multi-agent engine for writing full-length books

Android

Writing

Artificial Intelligence

Visit WebsiteSee on Product HuntPlay Store ⧉

Hello Product Hunt, I am the developers behind Scriptirix AI, and I are glad to share our work with you today. Most standard AI writing assistants eventually encounter a structural wall. When generating long-form manuscripts—whether a 150-page technical playbook, a complex business strategy guide, or an intricate novel—single-prompt systems quickly lose track of minor details. They drop technical variables, contradict previously defined business acronyms, leave plot holes, and introduce repetitive, generic prose. I built Scriptirix AI to approach long-form drafting as an engineered system, replacing the single-prompt model with an autonomous, multi-agent assembly line. 1. The Multi-Agent Assembly Line (Powered by Google Gemini) Instead of forcing one model to do everything, Scriptirix coordinates a specialized network of AI agents operating in parallel and sequential pipelines: Specialist Auditors: Depending on your book type, dedicated expert agents (such as the ROI Strategist, Technical Writer, Prose Stylist, or Subtext Injector) analyze the chapter brief first. They do not write the content; they generate highly precise, actionable structural and stylistic directives for the scene. The Head Orchestrator: This agent receives the combined, weighted expert reports and synthesizes them into high-fidelity prose, ensuring all technical and narrative guidelines are carried out. Self-Correcting QA (Lead Editor): The draft is immediately audited by a separate Gemini 3.1 Pro editor. It grades the prose based on a calibrated rubric covering pacing, logic, and style. If it detects sags, clichés, or logical contradictions, it generates a prescriptive list of defects and forces the writing agents to execute a surgical rewrite. 2. Surgical Remembrance of Micro-Details (No Plot Holes, No Logic Gaps) To prevent context drift and structural inconsistencies across multiple chapters, Scriptirix utilizes two distinct memory layers: Dynamic Continuity State Tracker: At the end of every chapter, a lightweight model extracts a "Delta State" mapping active developments. For non-fiction, it tracks newly defined acronyms, active corporate case studies, or mathematical variables. For fiction, it tracks unresolved mysteries, active tensions, and character inventories. This structured JSON state is preserved and injected directly into the context window of all subsequent runs. Vector RAG Memory: Completed chapters and raw reference files (such as custom .txt or .md source documents) are chunked, embedded using Google's text-embedding-001 model, and stored in a local pgvector database. When writing Chapter 20, the engine queries this database to recall exactly what was established in Chapter 2, eliminating logical drift. 3. Built for Non-Fiction first, Fine-Tuned for Fiction Scriptirix AI is optimized to handle different structural frameworks natively: Non-Fiction & Business Strategy (Primary): Tailored specifically for Technical Deep Dives, Business Playbooks, and Academic Research. The engine maintains rigorous factual adherence, handles code-snippet insertion cleanly, preserves complex business frameworks, and integrates research references without duplicating definitions. High-Fidelity Novels (Secondary): Configured for deep character development, world-building rules, natural dialogue flow, and subtext injection to ensure creative manuscripts retain narrative tension from the first page to the last. 4. Native Jetpack Compose Interface You can monitor the entire agent workflow in real-time. Our Android client streams the inner reasoning and handoffs of each specialist via WebSockets directly to a clean, "Liquid Obsidian" terminal interface. Because we believe in human-in-the-loop writing, you can manually edit book blueprints and chapter text in raw Markdown at any point, and the system will automatically sync the changes to your Google Drive. I would love to hear your thoughts, feedback, and technical questions about our multi-agent state tracking in the comments below!