Changelog

Switching into Split view, or leaving it for Preview, WYSIWYG or Raw, closed the application: the window reloaded. That is fixed. The toolbar also finds the editor again after a mode change, instead of going deaf.

A release that puts the house in order: ten menus that tell the truth about what the application does, a single command palette, and step-by-step assistants for the gestures that deserved guidance. Snapshots can finally be restored, and a defect that could leave them empty has been fixed.

A release entirely devoted to solidity: a full security audit followed by a cross-review combed through the application. Every path where text could be lost silently is now closed, reversible or reported.

The pack catalogue opens inside the application: ready-to-use expert methods to browse, buy and install without leaving the workshop — alongside a full visual refresh.

Anonymise a document without anything leaving your machine: local proper-name detection (NER), optional AI deep analysis, pseudonym replacement under your control.

Since then: this action is called Label identities in this file — one file, not a folder, no restore after rewriting. The gateway is Palimpseste Pseudo.

Model choice now follows the logical flow: connect, test, then pick among the models your server actually serves. And any server speaking the OpenAI API — LiteLLM gateway, vLLM, corporate proxy — becomes a first-class provider.

A complete methodology — personas, reference libraries, project template — now travels in a single .palimpack file. Two-click install, export from the File menu. Built for consulting firms and editorial teams distributing a ready-to-work environment.

Born from a full interface audit: contrast raised to WCAG AA, macOS "Reduce motion" honored, keyboard focus trapped on eight modals, unified editorial voice, export toasts with "Show in folder", last-session restore offered at startup, and a silent file backup before every AI operation.

Born from a full technical audit (no exploitable finding): atomic file writes everywhere — a crash mid-save can no longer truncate a file —, structural settings validation, anti-injection tests on the Markdown pipeline, offline license consistency check, and faster startup (main bundle down 36%). New built-in "Project audit & repair" guide in five languages.

For professionals producing structured chapter-by-chapter reports — audits, assessments, studies — Palimpseste removes the copy-paste friction between the AI chat and project files. A nine-chapter report now takes under thirty minutes of handling, down from roughly four hours.

Every sidebar panel — and the AI chat — can now go fullscreen with one click. When the Story Engine produces 1,500 words of prose, the split view felt cramped. Escape restores it.

Twenty classic narrative structures inject into your novel's timeline in one click. And sagas find their home: series mode manages several volumes under one roof, with a shared Bible and cross-volume continuity checks.

Printing (Cmd+P) now routes through the system browser — reliable on all three platforms, with access to every printer and native PDF export. Three vulnerabilities in the Mermaid diagram engine fixed. A discreet backup reminder on first launch.

The Story Engine preflight dialog now displays in the interface language.

The full-chapter generator, two-pass (beat-by-beat plan then markdown prose), 100% local-capable (Ollama, LM Studio) or cloud (Claude, GPT). Reads your Story Bible, plot threads, persona and document RAG. Preserves locked blocks and section anchors. Palimpseste's hammer feature.

Switch personas mid-project via a dedicated modal. Global Cmd+Alt+R "refresh-and-repair" shortcut to purge caches and rebuild the index without restarting the app. More robust migration of project files from older schemas.

The attached persona is now consumed by 4 AI surfaces (5 verbs + chat + 3 regens). Local semantic RAG on persona documents via fastembed-rs. User-scoped, cross-profile Reference Libraries attachable in parallel with the persona.

Full overhaul of the new-project flow: 4 cards (blank, template, import, library), 12 work templates, and introduction of Personas — your author voice captured locally, attached to a project in one click. Extensible project.yml audit with auto-fixers.

Local MoE models (Qwen 2.5 32B, Gemma 3 27B) sometimes fail on structured output. Palimpseste now tries 4 strategies in cascade: native → cooldown → channel-stripped → silent cloud fallback (opt-in). No more name-based blacklist to maintain.

100% local semantic AI memory (no cloud, no API key), per-profile workspaces, multi-model routing (one model per task), and auto-injected story bible for novelists. The AI gains depth without third-party dependencies.

Security hardening (signing-key rotation, automatic secret scan, dependency audit), explicit project-file versioning, full GDPR compliance, and bundle splitting for faster startup.

Automatic detection of your project's language and strict enforcement on AI models — your local assistant no longer drifts to English against your wishes.

Immediate diagnosis of incompatible models + tolerance window extended to 30 minutes for regeneration operations.

Import existing manuscripts from Word, PDF, HTML, RTF or Markdown in just a few clicks — automatic chapter detection, optional AI generation of the story bible and structure.

Transparent tracking of every AI-generated passage, permanent “100 % human” or “🤖 X % AI” indicator, and a new quality panel with AI marker detection and self-plagiarism check.