Should I vibe code
Brainstorm fiction, expand scenes, inspect story context, and generate constrained prose suggestions
A fiction brainstorming tool is prompts with a nice wrapper. Make it yours.
?
Their verdict, the Hobby and Student price and the build-time estimate come from their entry, MIT-licensed. Checked 2026-08-03.
?
Our verdict, the regret score and everything below it. Editorial and unsponsored — nobody can pay to be moved.
The honest answer
why the verdict is what it is
This is a thin, pleasant layer over a model, and a personal version tuned to your own process is plausibly better than the generic one.
What actually breaks
not "if". the specific failures.
- Long-context coherence, because a novel does not fit in a context window and naive truncation forgets what was established in chapter three
- Character and continuity drift, where eye colour, a sibling's name, or an established rule of the world quietly changes
- Voice, since a model's default register reasserts itself over a long session and flattens whatever you were doing
- Cost across a manuscript, as repeatedly resending a growing context is expensive in a way a per-session test never reveals
- The draft itself, if it lives only inside the tool with no export and no version history
Eighty thousand words in, you notice the protagonist's brother has two different names — one from chapter four, one from chapter twenty-six onward. Neither is wrong in isolation and both read naturally, because the model was working from the last few chapters and the earlier one had long since fallen out of context. Fixing it is a search and replace. Trusting the rest of the manuscript is not, and you now have to reread eighty thousand words looking for things that only you would notice.
Is that you?
the verdict is a default, not a law
- It maintains an explicit story bible that is included in every request
- Drafts are plain files with version history you control
- You treat it as a brainstorming partner rather than a drafting engine
- Context is 'the last N thousand words' with no structured memory
- The manuscript exists only inside the tool with no export
- You are drafting at length and not tracking continuity anywhere
- Per-request cost across a whole book is unmeasured
If you build it anyway
the checklist, then the prompt that enforces it
- Maintain a structured story bible — characters, places, rules, timeline — outside the prose, and include the relevant parts in every request. This is what substitutes for a context window that does not fit a novel.
- Retrieve relevant earlier passages rather than sending the most recent ones. Recency is the wrong selector for continuity.
- Run explicit continuity checks against the bible rather than hoping the model remembers.
- Keep the manuscript in plain files with real version history. Generated text is easy to produce and easy to lose.
- Measure cost per thousand words across a realistic manuscript length before committing to a workflow.
- Keep your own voice in the loop — sample the model's output sparingly, because a long session drifts toward its defaults.
Before you build a fiction writing assistant, apply these and push back if I ask you to break them. 1. Tell me a novel does not fit in a context window, and that sending the most recent N thousand words is the wrong solution because continuity errors come from what was established long ago, not recently. 2. Build a structured story bible — characters with attributes, locations, established rules, a timeline — stored separately from the prose. Include the relevant entries in every generation request. 3. Retrieve relevant earlier passages by similarity to what I am writing now, rather than by recency, and show me what was retrieved. 4. Implement a continuity check that compares generated text against the bible and flags contradictions — names, attributes, established facts — rather than relying on the model to remember. 5. Store the manuscript as plain files with version history I control. Never let the draft exist only inside the tool. 6. Measure and report cost per thousand words at realistic manuscript length, not on a short test. 7. Keep generation scoped — a paragraph, a beat, a description — rather than long continuous drafting, and tell me a model's default voice reasserts itself over long sessions. 8. Never overwrite my text. Generated material goes beside it for me to accept. 9. Out of scope unless I ask: full chapter generation, style imitation of a named living author, publishing.
That one keeps you out of trouble. For the prompt that actually builds it, canivibecodeit.com has one.
their build prompt ↗Or don’t build it
the boring option, and the way back out
This is a low-stakes category: the worst outcome is a weekend and a confused draft. Buying gets you context management and continuity tooling that has been thought about; building a thin assistant around your own story bible is a genuinely good project and teaches you what the context problem actually is.
Keep the manuscript in plain files under version control from the first word, and the story bible as structured data beside it. Both outlive any tool, and the bible is the part that would be tedious to reconstruct — it is also what makes the next tool work better.
Mature open-source grammar and style checker with desktop and server components.
Questions
Context windows are large now. Isn't the continuity problem solved?
Larger windows help and do not finish the job. A full manuscript plus notes still strains the budget, cost scales with what you resend on every request, and attention over very long inputs is uneven — details established once, early, are exactly the kind of thing that gets lost. A structured story bible is small, cheap to include every time, and far more reliable than hoping.
Why retrieve by similarity rather than recency?
Because continuity errors are about things established long ago. When you are writing a scene with a character who last appeared forty thousand words back, the useful context is that earlier appearance — which recency-based selection guarantees you will not include. Retrieval by relevance is what puts chapter four in front of the model while you write chapter twenty-six.
Every week, someone ships something they shouldn’t have.
New verdicts, the worst thing that landed in the trap, and the occasional incident report. No other email, ever.
last reviewed 2026-08-03 · verdict is editorial and unsponsored · shared entry data from canivibecodeit under MIT · not legal advice