Chekhov — AI editorial analysis for your manuscripts

Latte Other Software & Development
1

Hey everyone,

As a writer, I’m constantly hitting a wall I suspect most writers are familiar with: reviewing my manuscript — usually something feels off with some of my characters dialogue, or I forget to actually resolve a subplot, or any other kind of inconsistency that only storytellers know about.

I went looking for a tool that would read the whole manuscript and give me an editorial read, just like a real editor would do, but before submitting my manuscript to a real editorial analysis — I didn’t want the tool to write for me, not polish my prose, just tell me what was actually there. Couldn’t find one.

So I built it. It’s called Chekhov.

What Chekhov does:

It makes and actual AI Editorial Analysis for entire manuscripts (from first drafts on):

  • Character consistency — do people behave and speak like they are supposed to across the whole book?
  • Plot threads — what you opened, what you resolved, what quietly disappeared
  • Dialogue voice — does each character actually sound like a different person?
  • Timeline — maps it and flags contradictions

How it works with Scrivener:

On macOS and Windows, you import the .scriv project directly (The scriv file on macOS, and the scriv folder on Windows). Chekhov reads the binder structure as-is, so your chapters and scenes come through exactly as you’ve organized them. If you’ve built out character sheets in Scrivener’s Characters folder, those get picked up too and used to anchor the analysis.

Scrivener 2 and 3 files are both supported.

Then, Chekhov returns an idea of what your novel’s characters are. If it misses one, you can add it manually.

Finally, you run the analysis you want and let the magic happen!

Privacy first!

We don’t store your manuscript in any server. The only moment it leaves your computer is when sent to the AI model for the analysis, and that’s it. The rest of the time, your manuscript and every analysis you run are only stored in your own computer.

Pricing:

Analyses come in three depths — Standard (core editorial pass), Deep (adds dialogue voice and prose feedback), and Structural (the full pass, with pacing and themes). They cost 1, 2, and 3 credits respectively.

First Standard is free, no card needed. After that:

  • $5 for one credit
  • $12 for three
  • $20 for five.

No subscription.

Before you download anything, I put two sample reports on the site — Metamorphosis (Standard) and Wuthering Heights (Deep) — so you can see exactly what comes out of an analysis.

So, if you want to try it, I’d genuinely love to hear what you think — what worked, what didn’t, what you wish it did. I’m only one person behind this project, and the next round of features is going to come straight from what real writers tell me they need.

You can find more information, and the Download link of the app in the official website:

2

I don’t understand your pricing model. How is “Standard” free and costs 1 credit = $5 at the same time? “Deep” costs 2 credits, but you can buy just one or three credits. So you buy 2 x 1 credit = $10? “Structured” costs 3 credits = $12, that one makes sense. But if 3 credits is the highest tier, why is there a five credits = $20 option? :thinking:

3

Standard costs 1 credit BUT upon signing in for the first time you’ll get 1 credit for free, so you can run your first Standard Analysis and test the tool for yourself. Any other Standard Analysis will require you to purchase a credit.

The rest is just a bundle. If you wanted to run a Deep Analysis you can either purchase two credits or a bundle of 3 for 12usd. This makes way more sense if you wanted to run a Structural Analysis, which costs 3 credits. In that case, instead of purchasing 3 credits separately (which would cost you 15usd in total), you can buy the bundle for 12usd, saving 3usd in the process.

Does that make sense?

4

Since there is just one “Standard” Analysis, what you do mean by any other?

And what’s the five credits option for? Since there’s no plan that costs more than three credits or $12… Okay, so you pay $20 instead of $25, but you could just stop at $12 anyways, save $8 and still get the highest tier.

5

Since there is just one “Standard” Analysis, what you do mean by any other?

You can run standard analyses on as many manuscripts as you would like. So you could be running analyses on different manuscripts, or the same manuscript after changing it. So for example, you make your first draft, you analyze it with Chekhov, you get some issues to work on. You make a second draft. If you wanted to analyze that second draft, that would be a second Standard Analysis to run.

And what’s the five credits option for? Since there’s no plan that costs more than three credits or $12…

That’s actually a fair point, but as I said before, this tool is meant to work for more than one manuscript if a user needed it. So even if it’s correct that most users won’t be needing that 5 credit bundle all the time, possibly never, some users might, if they’re working on more than one manuscript at a time, for example, or want to run, let’s say, a Deep and Structural analyses after their first free Standard one.

So yes, it might not be quite useful for some people, but I wouldn’t mind leaving it there just in case.

6

Also, you wrote

That’s why Chekhov runs on your machine only, so your manuscript never leaves your side.

But your website says

An internet connection is required only when running an analysis.

What does it do with that connection? Since the manuscript never leaves your computer.

7

Yes. I need to make that clearer.

What I mean with privacy first is that I (Chekhov) don’t store any manuscript anywhere. You manuscript resides only in your own desktop app all the time. But there’s a slight exception: the analysis itself!

The analysis run on Anthropic’s Claude AI model. That’s on the internet. So for me to actually use Claude to analyze your manuscript I have to send it over to the Anthropic API for analysis. I don’t do anything else with your manuscripts. Only parse it to Anthropic for analysis and then return the result to your desktop app.

8

That’s the point I don’t understand. If you pay $5 for the Standard analysis (which you can run as often as you like), why would you pay $5 again for more standard analyes?

9

Nono wait. You don’t pay for an analysis per se. You pay for credits only. You buy credits and then, afterwards, do whatever you like with those credits. So if you purchase 5 credits you can either:

  • Run 5 different Standard Analyses.
  • Run 1 Deep Analysis and 1 Structural analysis
  • 1 Structural and 2 Standard

Or any other combination you want. They’re your credits. Use them however fits you.

10

If an analysis only runs if you buy credits first (every single time after the trial run?), how is buying credits not paying for the analysis?

And what else can I do with those credits? Other than buying more analyses?

With all that information in mind, and circling back to your first post…

Let’s see if I got it sorted out: My script never leaves my computer, except when your software performs its main function, and for that purpose I have to hand it to a person I don’t trust (nothing personal, I just don’t know you), so that you can hand it for me to a third party I absolutely don’t trust, who somehow works with it without storing it on their side. And I have to pay each time I want to run an analysis. Or spend the credits on ice cream. Is this correct?

3 Likes
11

If an analysis only runs if you buy credits first (every single time after the trial run?), how is buying credits not paying for the analysis?

It might as well be. I was just answering your first question:

If you pay $5 for the Standard analysis (which you can run as often as you like), why would you pay $5 again for more standard analyes?

You wouldn’t pay $5 for a standard analysis, you would only need to pay $1. That’s what I meant by: you purchase credits, not analyses. But sure, with those credits the only thing you can buy in the app are analyses.

My script never leaves my computer, except when your software performs its main function, and for that purpose I have to hand it to a person I don’t trust (nothing personal, I just don’t know you), so that you can hand it for me to a third party I absolutely don’t trust

Yes, that’s right. It’s the only secure way to handle this architecture. I understand your mistrust, but if you say you definitely don’t trust an AI vendor like Anthropic (which may be totally reasonable), then this tool just isn’t for you. Neither Chekhov nor any other tool will satisfy you because they all would need to send your manuscript to an AI. It’s just the way it is. I’m just trying to make the whole thing as safe as possible. Whatever Claude does with your manuscript is beyond my reach, same way if you decide to send portions of the same manuscript to ChatGPT by yourself.

12

Well, you can run an “A.I.” locally, without Internet connection. That’s a thing.

How are you making the whole process as safe as possible? If you have no control over what happens with the data after you sent it for me…

I have a hard time wrapping my head around this concept. You didn’t feel safe handing over your manuscript “to strangers”, who do whatever they want with it, and that inspired you to build an app that hands over your and other people’s manuscripts to a third party.

What has changed? Or what does your software to add more security, that you now feel better / safer about the whole procedure. I have a feeling I’m missing an important part.

3 Likes
14

“Only” is doing a lot of work there. Do you have any ability to ensure that Claude doesn’t hold onto the manuscript?

Also, what’s the specific value proposition that you offer? If Claude is doing the actual analysis, then what you’re offering is just a custom prompt, right? So why not just sell that and let the user deal with Anthropic themselves?

5 Likes
15

@farriagada I fear that @November_Sierra has tripped onto something that will give you serious troubles with any attempts to sell your service in the EU.

There is a common tactic in “freemium games” where by the in-game currency is only offered in inefficient bundles, so that you have to buy more than you need to get what you want from the in-game store.

Regulations are making this illegal in the EU. It is my understanding that once the enforcement date hits, you will have to show the exact price for every time in the ‘store’ and accept the exact price for every item at time of purchase. I am not a lawyer, and I have not read the final version of what the EU is working on, but this menu of token prices may cause issues.

I think I understand what you are trying to do, but I fear what you are actually doing is setting yourself up for a lot of paranoid carbon-based writers to look at you as an innocent agent that is in effect selling their work to be scraped by ChatGPA[1], Claude (or which cloud-based LLM you are using) and that their data – once it leaves YOUR figurative hands – will be treated in ways they neither understand nor likely to approve. As the person that your customers can see, you’ll catch the spears for the behavior of your service provider. You sure you want that much smoke?

That said…. There are three IF’s that would set my mind at ease to use an Analytical AI agent on my work. They are:

IF – my work never leaves my machine, neither as raw manuscript nor as tokenized data.[2]

IF – I retain full control on the output of the process and that only ancillary usage information gets handed back to the developer.[3]

IF – I can set the analysis to run on my schedule with an airgapped machine, not connected to the internet at the time of analysis.[4]

All that said, I’d like a tool that I could pass my project file to and get analysis from, but like most writers out there, I am fearful that an AI be trained on my work, and that my work will end up feathering the nest of someone that doesn’t need the extra money, and is already responsible for so much strife in computing right now with the monopolistic globbing of RAM, storage, C/A/Gpu chips, and grid power.

Sidebar: Have you tried one of the Open AI models (like Ollama) that run on local hardware?

For most of us writers out there that can spend hours at a whack writing, pretending to write while checking forums, or even months on crafting novels, we can wait a little longer for the return of reports, so not having a vast cloud-computing array to crunch the numbers isn’t really a turn off.

That said, I see potential in this project of yours, if you can get past the teething pains.

Tell Pavel I said hi.

  1. My nickname for ChatGPT as it seems to be most commonly used for making one appear more intellgent than they are, just like a 4.0 GPA makes one appear more smart than they might actually be. ↩︎

  2. I have made my peace with Dropbox and iCloud and similar STORAGE systems. That said, I don’t write with Google Docs for a reason, because I don’t trust that system not to leverage my data for their profit. ↩︎

  3. Things like how many times I ran which analysis are fine, which features I used and which I never use, all fine. Things like statistical measurements of how many ‘-ly’ words I use, or other similar language foilbles are iffy. Which programs are active, what cookies are in my browser cache, and similar invasive stuff is not, ↩︎

  4. To be clear here, I mean I can take my MacBook to the mountaintop or the far side of the moon and get a report run. I can stomach ‘phone home’ license checks if they are fortnight infrequent but refuse to use anything that requires a permanent and active internet connection to function. It’s practical reasons as much as anything else because as vast as cellular coverage is in North America, I still find pockets and places where there isn’t usable mobile internet, and frankly those sorts of places are often where truck parking is to be found. ↩︎

4 Likes
16

I think you may be misunderstanding your target market here. I don’t know any writers that would trust this, want this, or need this. I know I don’t. I wouldn’t give them my work directly let alone give it to a third party/middle man to do it for me. Not even for free, let alone for a fee. I actively go out of my way to avoid any service that uses AI or LLM of any kind.

9 Likes
17

I use Claude extensively every day at work (software engineering/code/tooling - both using their apps/integrations and by pushing things to the Anthropic APIs). As a writer, I’d be extremely cautious putting my writing anywhere near it: This is its own ambiguous response when asked…

Screenshot 2026-04-22 at 11.49.15
Screenshot 2026-04-22 at 11.49.151398×240 37.1 KB

Then, the perpetual elephant in the room: did Anthropic pay for the content used for training (we know the answer to that…)

I don’t mean to rain on an interesting project, but it’s important to be clear about how these things work, how they were built, and the risks.

If you are wrapping Claude in a service, these privacy/settings also become your responsibility.

5 Likes
18

I’ll be brief: due to the inherently probabilistic nature of LLMs, any artificial analyst will tend to evaluate works according to criteria of normality that inevitably lead to mediocrity.

Furthermore, these models do not work well (and potentially will not) in languages ​​other than English.

Regards!

8 Likes
19

Exactly why depending on these tools to “fix” clunky phrasing is such a terrible idea. If you let an LLM airbrush away the quirks that make your writing unique, what’s left?

9 Likes
20

Tennessee Williams once gave Gore Vidal a story of his to read. Vidal didn’t like it and offered to improve it: ‘I reversed backward-running sentences, removed repetitions, eliminated half those adjectives and adverbs he always insisted do their work in pairs. I was proud of the result. He was deeply irritated. “What you have done is remove my style, which is all that I have.”’

A

5 Likes