Open-source CLI AI Agent that autonomously writes, audits, and revises novels — with human review gates that keep you in control. Supports LitRPG, Progression Fantasy, Isekai, Romantasy, Sci-Fi, and more. Continuation, spinoff, fanfic, and style imitation workflows built in.
InkOS Studio is here! — run inkos studio to launch a local web workbench. Book management, chapter review & editing, real-time writing progress, market radar, analytics, AI detection, style analysis, genre management, daemon control, truth file editing — everything the CLI does, now visual.
Native English novel writing now supported！ — 10 built-in English genre profiles with dedicated pacing rules, fatigue word lists, and audit dimensions. Set --lang en and go.
npm i -g @actalk/inkosInkOS is published as an OpenClaw Skill, callable by any compatible agent (Claude Code, OpenClaw, etc.):
clawhub install inkos # Install from ClawHubIf you installed via npm or cloned the repo, skills/SKILL.md is already included — 🦞 can read it directly without a separate ClawHub install.
Once installed, Claw can invoke InkOS atomic commands and control-surface operations (plan chapter/compose chapter/draft/audit/revise/write next) via exec, with --json output for structured decision-making. The recommended flow is: update author_intent.md or current_focus.md, run plan / compose, then decide whether to call draft or the full write next pipeline. You can also browse it on ClawHub by searching inkos.
Option 1: Global config (recommended, one-time setup)
inkos config set-global \
--lang en \
--provider <openai|anthropic|custom> \
--base-url <API endpoint> \
--api-key <your API key> \
--model <model name>
# provider: openai / anthropic / custom (use custom for OpenAI-compatible proxies)
# base-url: your API provider URL
# api-key: your API key
# model: your model name--lang en sets English as the default writing language for all projects. Saved to ~/.inkos/.env. New projects just work without extra config.
Option 2: Per-project .env
inkos init my-novel # Initialize project
# Edit my-novel/.env# Required
INKOS_LLM_PROVIDER= # openai / anthropic / custom (use custom for any OpenAI-compatible API)
INKOS_LLM_BASE_URL= # API endpoint
INKOS_LLM_API_KEY= # API Key
INKOS_LLM_MODEL= # Model name
# Language (defaults to global setting or genre default)
# INKOS_DEFAULT_LANGUAGE=en # en or zh
# Optional
# INKOS_LLM_TEMPERATURE=0.7 # Temperature
# INKOS_LLM_MAX_TOKENS=8192 # Max output tokens
# INKOS_LLM_THINKING_BUDGET=0 # Anthropic extended thinking budgetProject .env overrides global config. Skip it if no override needed.
Option 3: Multi-model routing (optional)
Assign different models to different agents — balance quality and cost:
# Assign different models/providers to different agents
inkos config set-model writer <model> --provider <provider> --base-url <url> --api-key-env <ENV_VAR>
inkos config set-model auditor <model> --provider <provider>
inkos config show-models # View current routingAgents without explicit overrides fall back to the global model.
Structured State + Hook Governance + Length Governance
Addresses three systemic long-form writing problems: context bloat after 20+ chapters causing slowdowns and 400 errors (Settler full injection → JSON delta + selective retrieval), hooks only accumulate, never resolve, ~0% payoff rate (Planner scheduling + Settler blind spot fix + audit debt tracking), word count deviation 50%+ and normalizer destroying chapters (LengthSpec + safety net).
- Pipeline upgraded to 10 agents: adds Planner, Composer, Observer, Reflector, Normalizer
- Truth files moved to
story/state/*.json(Zod validated); Settler outputs JSON delta instead of full markdown; legacy books auto-migrate - SQLite temporal memory database on Node 22+ for relevance-based retrieval
- Planner generates
hookAgendato schedule hook advancement and payoff; Settler working set expanded to cover dormant debt - New
mentionsemantics prevents fake hook advancement;analyzeHookHealthaudits hook debt;evaluateHookAdmissionblocks duplicate hooks - Length governance:
LengthSpec+ Normalizer single-pass correction with safety net against destructive normalization - User
INKOS_LLM_MAX_TOKENSacts as global cap; reserved keys inllm.extraauto-stripped - Cross-chapter repetition detection, dialogue-driven guidance, English variance brief, multi-character scene resistance
- Chapter summary dedup, ESM node:sqlite fix, consolidate full-width parenthesis support
- Bilingual CLI output and logging
English is the default for English genre profiles. Pick a genre and go:
inkos book create --title "The Last Delver" --genre litrpg # LitRPG novel (English by default)
inkos write next my-book # Write next chapter (full pipeline: draft → audit → revise)
inkos status # Check status
inkos review list my-book # Review drafts
inkos review approve-all my-book # Batch approve
inkos export my-book --format epub # Export EPUB (read on phone/Kindle)Language is set per-genre by default. Override explicitly with --lang en or --lang zh. Use inkos genre list to see all available genres and their default languages.
InkOS ships with 10 English-native genre profiles. Each includes genre-specific rules, pacing, fatigue word detection, and audit dimensions:
| Genre | Key Mechanics |
|---|---|
| LitRPG | Numerical system, power scaling, stat progression |
| Progression Fantasy | Power scaling, no numerical system required |
| Isekai | Era research, world contrast, cultural fish-out-of-water |
| Cultivation | Power scaling, realm progression |
| System Apocalypse | Numerical system, survival mechanics |
| Dungeon Core | Numerical system, power scaling, territory management |
| Romantasy | Emotional arcs, dual POV pacing |
| Sci-Fi | Era research, tech consistency |
| Tower Climber | Numerical system, floor progression |
| Cozy Fantasy | Low-stakes pacing, comfort-first tone |
Also supports 5 Chinese web novel genres (xuanhuan, xianxia, urban, horror, other) for bilingual creators.
Every genre includes a fatigue word list (e.g., "delve", "tapestry", "testament", "intricate", "pivotal" for LitRPG) — the auditor flags these automatically so your prose doesn't read like every other AI-generated novel.
The Continuity Auditor agent checks every draft across 33 dimensions: character memory, resource continuity, hook payoff, outline adherence, narrative pacing, emotional arcs, and more. Built-in AI-tell detection automatically catches "LLM voice" — overused words, monotonous sentence patterns, excessive summarization. Failed audits trigger an automatic revision loop.
De-AI-ification rules are baked into the Writer agent's prompts: fatigue word lists, banned patterns, style fingerprint injection — reducing AI traces at the source. revise --mode anti-detect runs dedicated anti-detection rewriting on existing chapters.
inkos style analyze examines reference text and extracts a statistical fingerprint (sentence length distribution, word frequency patterns, rhythm profiles) plus an LLM-readable style guide. inkos style import injects this fingerprint into a book — all future chapters adopt the style, and the Reviser audits against it.
inkos book create --brief my-ideas.md — pass your brainstorming notes, worldbuilding doc, or character sheets. The Architect agent builds from your brief (generating story_bible.md and book_rules.md) instead of inventing from scratch, and persists the brief into story/author_intent.md so the book's long-horizon intent does not disappear after initialization.
Every book now has two long-lived Markdown control docs:
story/author_intent.md: what this book should become over the long horizonstory/current_focus.md: what the next 1-3 chapters should pull attention back toward
Before writing, you can run:
inkos plan chapter my-book --context "Pull attention back to the mentor conflict first"
inkos compose chapter my-bookThis generates story/runtime/chapter-XXXX.intent.md, context.json, rule-stack.yaml, and trace.json. intent.md is the human-readable contract; the others are execution/debug artifacts. plan / compose only compile local documents and state, so they can run before you finish API key setup.
draft, write next, and revise now share the same conservative length governor:
--wordssets a target band, not an exact hard promise- Chinese chapters default to
zh_chars; English chapters default toen_words - If the chapter drifts outside the soft band, InkOS may run one corrective normalization pass (compress or expand) instead of hard-cutting prose
- If the chapter still misses the hard range after that one pass, InkOS still saves it, but surfaces a visible length warning and telemetry in the result and chapter index
inkos import chapters imports existing novel text, auto reverse-engineers all 7 truth files (world state, character matrix, resource ledger, plot hooks, etc.), supports Chapter N and custom split patterns, and resumable import. After import, inkos write next seamlessly continues the story.
inkos fanfic init --from source.txt --mode canon creates a fanfic book from source material. Four modes: canon (faithful continuation), au (alternate universe), ooc (out of character), cp (ship-focused). Includes a canon importer, fanfic-specific audit dimensions, and information boundary controls to keep lore consistent.
Different agents can use different models and providers. Writer on Claude (stronger creative), Auditor on GPT-4o (cheaper and fast), Radar on a local model (zero cost). inkos config set-model configures per-agent; unconfigured agents fall back to the global model.
inkos up starts an autonomous background loop that writes chapters on a schedule. The pipeline runs fully unattended for non-critical issues, pausing for human review when needed. Notifications via Telegram and Webhook (HMAC-SHA256 signing + event filtering). Logs to inkos.log (JSON Lines), -q for quiet mode.
Supports any OpenAI-compatible endpoint (--provider custom). Stream auto-fallback — when SSE isn't supported, InkOS retries with sync mode automatically. Fallback parser handles non-standard output from smaller models, and partial content recovery kicks in on stream interruption.
Every chapter creates an automatic state snapshot — inkos write rewrite rolls back any chapter to its pre-write state. The Writer outputs a pre-write checklist (context scope, resources, pending hooks, risks) and a post-write settlement table; the Auditor cross-validates both. File locking prevents concurrent writes. Post-write validator includes cross-chapter repetition detection and 11 hard rules with auto spot-fix.
The hook system uses Zod schema validation — lastAdvancedChapter must be an integer, status can only be open/progressing/deferred/resolved. JSON deltas from the LLM are processed through applyRuntimeStateDelta (immutable update) and validateRuntimeState (structural check) before persistence. Corrupted data is rejected, not propagated.
User-configured INKOS_LLM_MAX_TOKENS now acts as a global cap on all API calls. Reserved keys in llm.extra (max_tokens, temperature, etc.) are automatically stripped to prevent accidental overrides.
Each chapter is produced by multiple agents in sequence, with zero human intervention:
| Agent | Responsibility |
|---|---|
| Radar | Scans platform trends and reader preferences to inform story direction (pluggable, skippable) |
| Planner | Reads author intent + current focus + memory retrieval results, produces chapter intent (must-keep / must-avoid) |
| Composer | Selects relevant context from all truth files by relevance, compiles rule stack and runtime artifacts |
| Architect | Plans chapter structure: outline, scene beats, pacing targets |
| Writer | Produces prose from the composed context (length-governed, dialogue-driven) |
| Observer | Over-extracts 9 categories of facts from the chapter text (characters, locations, resources, relationships, emotions, information, hooks, time, physical state) |
| Reflector | Outputs a JSON delta (not full markdown); code-layer applies Zod schema validation then immutable write |
| Normalizer | Single-pass compress/expand to bring chapter length into the target band |
| Continuity Auditor | Validates the draft against 7 canonical truth files, 33-dimension check |
| Reviser | Fixes issues found by the auditor — auto-fixes critical problems, flags others for human review |
If the audit fails, the pipeline automatically enters a revise → re-audit loop until all critical issues are resolved.
Every book maintains 7 truth files as the single source of truth:
| File | Purpose |
|---|---|
current_state.md |
World state: character locations, relationships, knowledge, emotional arcs |
particle_ledger.md |
Resource accounting: items, money, supplies with quantities and decay tracking |
pending_hooks.md |
Open plot threads: foreshadowing planted, promises to readers, unresolved conflicts |
chapter_summaries.md |
Per-chapter summaries: characters, key events, state changes, hook dynamics |
subplot_board.md |
Subplot progress board: A/B/C line status tracking |
emotional_arcs.md |
Emotional arcs: per-character emotion tracking and growth |
character_matrix.md |
Character interaction matrix: encounter records, information boundaries |
The Continuity Auditor checks every draft against these files. If a character "remembers" something they never witnessed, or pulls a weapon they lost two chapters ago, the auditor catches it.
Since 0.6.0, the authoritative source for truth files has moved from markdown to story/state/*.json (Zod schema validated). The Settler no longer outputs full markdown files — it produces a JSON delta that is immutably applied and structurally validated before persistence. Markdown files are retained as human-readable projections. Existing books auto-migrate on first run.
On Node 22+, a SQLite temporal memory database (story/memory.db) is automatically enabled, supporting relevance-based retrieval of historical facts, hooks, and chapter summaries — preventing context bloat from full-file injection.
Alongside the 7 truth files, InkOS splits guardrails from customization into reviewable control docs:
story/author_intent.md: long-horizon author intentstory/current_focus.md: near-term steeringstory/runtime/chapter-XXXX.intent.md: chapter goal, keep/avoid list, conflict resolutionstory/runtime/chapter-XXXX.context.json: the actual context selected for this chapterstory/runtime/chapter-XXXX.rule-stack.yaml: priority layers and override relationshipsstory/runtime/chapter-XXXX.trace.json: compilation trace for this chapter
That means briefs, outline nodes, book rules, and current requests are no longer mashed into one prompt blob; InkOS compiles them first, then writes.
The Writer agent has ~25 universal writing rules (character craft, narrative technique, logical consistency, language constraints, de-AI-ification), applicable to all genres.
On top of that, each genre has dedicated rules (prohibitions, language constraints, pacing, audit dimensions), and each book has its own book_rules.md (protagonist personality, numerical caps, custom prohibitions), story_bible.md (worldbuilding), author_intent.md (long-horizon direction), and current_focus.md (near-term steering). volume_outline.md still acts as the default plan, but in v2 input governance it no longer automatically overrides the current chapter intent.
InkOS provides three interaction modes, all sharing the same atomic operations:
inkos write next my-book # Draft → audit → auto-revise, all in one
inkos write next my-book --count 5 # Write 5 chapters in sequencewrite next now uses the plan -> compose -> write governance chain by default. If you need the older prompt-assembly path, set this explicitly in inkos.json:
{
"inputGovernanceMode": "legacy"
}The default is now v2. legacy remains available as an explicit fallback.
inkos plan chapter my-book --context "Focus on the mentor conflict first" --json
inkos compose chapter my-book --json
inkos draft my-book --context "Focus on the dungeon boss encounter and party dynamics" --json
inkos audit my-book 31 --json
inkos revise my-book 31 --jsonEach command performs a single operation independently. --json outputs structured data. plan / compose govern inputs; draft / audit / revise handle prose and quality checks. They can be called by external AI agents via exec, or used in scripts.
inkos agent "Write a LitRPG novel where the MC is a healer class in a dungeon world"
inkos agent "Write the next chapter, focus on the boss fight and loot distribution"
inkos agent "Create a progression fantasy about a mage who can only use one spell"18 built-in tools (write_draft, plan_chapter, compose_chapter, audit_chapter, revise_chapter, scan_market, create_book, update_author_intent, update_current_focus, get_book_status, read_truth_files, list_books, write_full_pipeline, web_fetch, import_style, import_canon, import_chapters, write_truth_file), with the LLM deciding call order via tool-use. The recommended agent flow is: adjust the control surface first, then plan / compose, then choose draft-only or full-pipeline writing.
| Command | Description |
|---|---|
inkos init [name] |
Initialize project (omit name to init current directory) |
inkos book create |
Create a new book (--genre, --chapter-words, --target-chapters, --brief <file>, --lang en/zh) |
inkos book update [id] |
Update book settings (--chapter-words, --target-chapters, --status, --lang) |
inkos book list |
List all books |
inkos book delete <id> |
Delete a book and all its data (--force to skip confirmation) |
inkos genre list/show/copy/create |
View, copy, or create genres |
inkos plan chapter [id] |
Generate the next chapter's intent.md (--context / --context-file for current steering) |
inkos compose chapter [id] |
Generate the next chapter's context.json, rule-stack.yaml, and trace.json |
inkos write next [id] |
Full pipeline: write next chapter (--words to override, --count for batch, -q quiet mode) |
inkos write rewrite [id] <n> |
Rewrite chapter N (restores state snapshot, --force to skip confirmation) |
inkos draft [id] |
Write draft only (--words to override word count, -q quiet mode) |
inkos audit [id] [n] |
Audit a specific chapter |
inkos revise [id] [n] |
Revise a specific chapter |
inkos agent <instruction> |
Natural language agent mode |
inkos review list [id] |
Review drafts |
inkos review approve-all [id] |
Batch approve |
inkos status [id] |
Project status |
inkos export [id] |
Export book (--format txt/md/epub, --output <path>, --approved-only) |
inkos fanfic init |
Create a fanfic book from source material (--from, --mode canon/au/ooc/cp) |
inkos config set-global |
Set global LLM config (~/.inkos/.env) |
inkos config set-model <agent> <model> |
Per-agent model override (--base-url, --provider, --api-key-env) |
inkos config show-models |
Show current model routing |
inkos doctor |
Diagnose setup issues (API connectivity test + provider compatibility hints) |
inkos detect [id] [n] |
AIGC detection (--all for all chapters, --stats for statistics) |
inkos style analyze <file> |
Analyze reference text to extract style fingerprint |
inkos style import <file> [id] |
Import style fingerprint into a book |
inkos import chapters [id] --from <path> |
Import existing chapters for continuation (--split, --resume-from) |
inkos analytics [id] / inkos stats [id] |
Book analytics (audit pass rate, top issues, chapter ranking, token usage) |
inkos studio |
Start web workbench (-p for port, default 4567) |
inkos up / down |
Start/stop daemon (-q quiet mode, auto-writes inkos.log) |
[id] is auto-detected when the project has only one book. All commands support --json for structured output. draft / write next / plan chapter / compose chapter accept --context for steering, and --words overrides the target chapter size. book create supports --brief <file> to pass a creative brief — the Architect builds from your ideas instead of generating from scratch. plan chapter / compose chapter do not require a live LLM, so you can inspect governed inputs before finishing API setup.
-
packages/studioWeb UI workbench (Vite + React + Hono)inkos studio - Interactive fiction (branching narrative + reader choices)
- Partial chapter intervention (rewrite half a chapter + cascade truth file updates)
- Custom agent plugin system
Contributions welcome. Open an issue or PR.
pnpm install
pnpm dev # Watch mode for all packages
pnpm test # Run tests
pnpm typecheck # Type-check without emitting