THE NOVELFLOW CHANGELOG

A little better,
every chapter.

Every version, from the very beginning. More room for the story. Less friction along the way.

Explore the user guide
LATEST UPDATETRIAL
0.2.2
Model setup recovery for first-time projects
v0.2.2Test build

Model setup recovery for first-time projects

When a fresh installation has no usable model yet, project setup keeps the author’s inputs and offers a clear, returnable path to configuration.

IN FOCUS · YOUR WRITING FLOW
  1. 01Gather context
  2. 02Draft a proposal
  3. 03Review & refine
  4. 04Author approves

Fixes

  • Generate Scope Candidates no longer exposes an internal error code and instead explains that a usable model is missing.
  • The premise, platform, target length, and selected references remain in place so the author can return after model setup.
  • Historical tasks still resume against their frozen model identity, preserving retry consistency checks.

Good to know

  • The public build is distributed as a Windows x64 installer through a signed update manifest; AI features require the author’s own model service.
v0.2.1Test build

A clearer workspace with resilient imports

Writing, workflow, and story-reference screens now share a consistent hierarchy, with safer recovery for mixed imports and older databases.

Improvements

  • The writing home, sidebars, workflow, audit trail, characters, factions, rules, and items now share a consistent control and information hierarchy.
  • Page changes preserve the loaded workspace and scroll position, with the writing footer, focus controls, statistics, and mode switch realigned.
  • Project setup exposes genre and platform choices directly, while source checks can review all constraints and save each result.
  • Scheduled graph extraction stays collapsed by default and starts, checkpoints, and stops on schedule while manual author actions retain priority.

Fixes

  • When one item conflicts during a mixed reference-and-manuscript import, only that item is isolated; the rest continues and successful chapters are not duplicated.
  • Characters, factions, rules, and items can be fully viewed, edited, and persisted with consistent source, manuscript location, and evidence boundaries.
  • When modifying an older installation exposes a volume-order conflict, startup creates a verifiable backup before a narrowly scoped idempotent repair.
  • Assistant failure messages are deduplicated, and saved whole-book outline candidates remain marked for confirmation and available to later context.

Good to know

  • About & Updates uses Gitee as the primary domestic channel and GitHub as a fallback, with both channels publishing the same signed package identity.
v0.2.0Test build

From setup to confirmed writing

Project setup, chapter evidence, workflow execution, and author confirmation now form one recoverable writing path, with a calmer workspace for long sessions.

New features

  • New projects include a guided setup flow; accepted setting candidates become editable reference entries with their source retained.
  • The chapter notebook gathers relevant settings, graph facts, memories, and manuscript evidence, with follow-up retrieval when evidence is missing.
  • The workflow library now offers complete examples, functional groups, and foundational modules that can be used whole or composed per task.
  • Installed builds add signed update checks, download, installation, and restart handling with a primary update source and a synchronized package source.

Improvements

  • Chat and the workflow canvas now share the same task, node progress, and recovery record, with each result passed explicitly to the next step.
  • Confirmed writes, persisted readback, and knowledge synchronization are shown as distinct stages so completion remains clear.
  • Writing sidebars can collapse and prose width can be adjusted, while the editor toolbar, import classification, and themed scrollbars use a tighter layout.

Fixes

  • Completed operations whose response is briefly lost recover under the original task identity, preventing duplicate writes and false failure states.
  • Recovery keeps generated candidates and task records, allowing knowledge synchronization to retry without regenerating completed work.
  • Long-form evidence preparation and continuity review use bounded retrieval and recovery requests instead of waiting indefinitely or treating failure as success.

Good to know

  • This test build was checked across packaged installers, a native window, and data integrity. AI features still require a compatible service configured by the user, and generated changes remain author-confirmed.
v0.1.10Test build

A more dependable writing agent

Confirmed agent actions can now manage workflows, skills, and selected writing preferences, while recovering completed work when a response is briefly lost.

New features

  • After confirmation, the agent can create or update workflows and skills, bind task flows, and complete character cards.
  • The agent can adjust non-sensitive preferences such as theme, manuscript font size, indentation, autosave delay, and automatic mirroring.
  • New projects include a reference chapter before the manuscript for worldbuilding, outlines, character biographies, and writing rules.

Improvements

  • Reference chapters remain separate from prose; graph extraction can use prose, references, or both while preserving their distinct evidence meaning.
  • Naturalness review shows stage progress and bounds waiting time; content is not written to the manuscript before review converges.
  • Workflow lists, default task bindings, and live progress now form one path, making switching and comparison more direct.
  • The Story Library, dual outlines, action cards, and model selector use a tighter hierarchy that remains usable in narrow windows.

Fixes

  • Committed actions whose response was briefly lost can be recovered by their original identity, preventing duplicates and false failures.
  • Workflow and skill writes check the current revision and read back persisted state, preventing stale data from overwriting newer edits.
  • Reference chapters no longer affect manuscript order, chapter numbering, batch generation, or prose counts in full-book review.

Good to know

  • This is a fully desktop-tested build; cloud AI still requires a compatible service configured by the user, and sensitive service settings remain unavailable to the agent.
v0.1.9Test build

Reference chapters and task flows

Creative references now have a home outside the manuscript, and each writing task can enter its assigned workflow.

New features

  • New projects include a reference chapter before the manuscript for worldbuilding, outlines, biographies, and rules, while still opening on the first prose chapter.
  • Graph extraction can explicitly use prose, references, or both while preserving source type and evidence boundaries.

Improvements

  • Diagnosis, direction, polishing, expansion, and chat can each use a default workflow, with progress shared by the writing panel and studio.
  • Project workflows can be created and switched from the left list, and the active workflow persists across restarts.

Fixes

  • Fixed reference chapters affecting manuscript order, generated chapter numbering, and counts, plus workflow revision conflicts after restart.
  • Graph preparation now uses bounded waiting and safe retry instead of hanging indefinitely on database problems.
v0.1.8Test build

Custom workflow studio

Imports can separate prose, outlines, and story references, while writers can assemble and compare their own visual AI workflows.

New features

  • The workflow canvas combines agent, context, skill, model, review, revision, and write modules, with save, copy, import, and export.
  • The studio shows the active node, stage description, elapsed time, completion progress, and token use while a workflow runs.
  • Prose, scores, elapsed time, and usage from the same input can be compared side by side.

Improvements

  • Source-aware import routes prose, four-level outlines, and story references to their own workspaces.
  • Projects containing only outlines or references can create a dedicated reference volume and feed the Story Library, information store, and graph.
  • Non-prose references can switch between Markdown editing and preview while manuscript chapters remain plain text.

Fixes

  • Fixed concurrent workflow initialization, restart recovery, author nodes lost during graph refresh, and controls colliding in narrow panels.
v0.1.7Test build

Author outlines beside AI references

Each chapter now keeps the author outline and AI reference separately, with safer graph rebuilding and writing-task recovery.

New features

  • The outline studio adds side-by-side Author Outline and AI Reference columns; both can be edited, saved, and read by the agent.

Improvements

  • Each chapter can extract an outline from saved prose or request a separate AI-generated reference outline.
  • Model configuration, native networking, and connection diagnostics share one entry point, while reasoning remains separate from final prose.

Fixes

  • Full graph refresh preserves people, places, relations, and evidence.
  • Writing tasks retain clear states during recovery, and completed steps are not repeated.

Good to know

  • The author outline remains the canonical story line, while AI-generated material stays an independent reference.
v0.1.6Test build

Long-form context and skills

Prose, outlines, references, graphs, and memory now share a visible context plan, alongside editable skills and confirmation-based agent writes.

New features

  • A context meter and source ledger show what the current run actually uses.
  • The skill store supports editing, saving, and restoring built-in skills, composed as base, conditional, and safety layers.
  • The writing agent supports continuation, chapter batches, full-book review, and proposed revisions.

Improvements

  • Large reference collections are parsed in batches, deduplicated, and retrieved as needed, while multilevel outlines remain in the outline studio.
  • Prose generation connects plot detail, continuity review, naturalness review, local revision, and independent final review.

Good to know

  • AI writes first appear as confirmation cards; service speed and availability depend on the user’s provider.
v0.1.5Test build

Traceable agent context

Writers can see which prose and references the AI used; proposed changes enter the work only after confirmation and support undo and recovery.

New features

  • A unified context ledger records source, revision, and capacity estimates.
  • Writing skills can be composed as base, conditional, and safety layers while retaining editable versions.
  • Proposed-action cards support confirmation, undo, recovery, and information-store synchronization.

Improvements

  • The story atlas adds regions, places, routes, chapter playback, base-map calibration, and safe backup with assets.
  • AI service settings support multiple profiles, connection tests, and user-authorized compatible endpoints.

Good to know

  • When the writer explicitly requests the full manuscript, chapters enter the final request separately and are checked before transmission.
v0.1.4Test build

Four-level story outlines

Structure from book to chapter outline can be edited, imported, and exported, then linked with prose, graphs, and reviewable AI actions.

New features

  • The plot workspace provides a book, volume, arc, and chapter-outline tree with moving, sorting, soft delete, recovery, and edit history.
  • Four-level outlines can be previewed and imported from Markdown, TXT, or NovelFlow JSON, then exported as Markdown or JSON.
  • Chapter outlines can link to prose using either per-volume or book-wide continuous numbering.
  • The story graph adds author-planned nodes, manuscript evidence, relation provenance, and a more spatial 3D view.
  • AI changes enter a review queue where writers can inspect context, accept, reject, undo, and continue after interruption.

Improvements

  • The graph distinguishes planned, candidate, verified, and conflicted states for checking references against prose.
  • Outlines, references, memory, AI actions, and undo records are saved as one consistent write.

Fixes

  • Fixed outline imports retaining old project chapter identifiers and linking across volumes when a same-volume chapter was unavailable.
  • Fixed imported results repeating volume names, arc names, and structural numbering.
v0.1.3Test build

More reliable long-form import

Long-form import now separates prose, volumes, and story material, with better database handling, model connections, and error feedback.

Improvements

  • The import flow better handles long files and routes manuscript content and story references to the appropriate places.
  • Volume detection and chapter organization are more stable, making imported work easier to continue editing.

Fixes

  • Improved handling of a busy database, reducing blocks during saves and project switching.
  • Model connection and import failures now provide clearer causes and recovery guidance.
v0.1.2Candidate

Story and memory integration candidate

This development candidate aligned the Story Library, plot outlines, memory, and AI context, but did not produce a separate user test package.

Improvements

  • Story references, plot structure, and memory candidates began feeding one writing context.

Good to know

  • The version reached application manifests and its changes later shipped through 0.1.3 and 0.1.4; this entry does not represent a standalone release.
v0.1.1Test build

Story Library and memory candidates

The core writing tool gained a Story Library, memory candidates, AI context, and fuller story-graph controls.

New features

  • Added structured management for characters, places, and other story references.
  • Added memory candidates for author review before information enters later writing context.

Improvements

  • The AI assistant can use more project context, reducing suggestions that drift from the story.
  • Graph nodes and relationships gain fuller maintenance controls.
v0.1.0Test build

The first limited test build

NovelFlow first brought projects, chapters, manuscript editing, AI writing assistance, and a story graph into one local Windows app.

New features

  • Writers can create local projects and organize volumes and chapters.
  • The manuscript editor supports local writing and saving.
  • The AI writing assistant can suggest changes using the current work.
  • The first story graph organizes characters, places, and their relationships.

Good to know

  • This build was used for limited testing and was not offered as a public download.

The story continues. Thanks for being part of it.

Share your feedback ↗