What you walk away with
You sit down with a brief and you get a draft that still looks like the piece you asked for. Draftwright is the writing desk for that. You choose how to start, you choose the model, and you read a draft that keeps the shape of the assignment. If you asked for numbered sections, they are still numbered. If you asked for a long piece, you do not get a fragment in its place. The version you approve is kept, so the next draft has an example of wording you will stand behind.
There are six ways to start. Ask, when you have a question. Create content, when the piece is meant to be published. Summarize, and Rewrite, when the material is already in front of you. Deep Research, when the draft should come from public pages and show you those pages. Website Study, when you want new copy drawn from one public site, and only that site. You pick the one that matches the job. You do not have to force every task through the same box.
The first draft uses the model you selected. The response can show you which model ran, which saved guidance was used, and whether the draft still matches the assignment. If the workflow gathered public sources, they come back with the draft. You are not left wondering which page a claim came from. You can approve the final wording, or dislike it and say why. Both decisions are there for the next piece.
What you walk away with is a draft you can edit, a record of the decision you made about it, and a desk that will use that decision again. The brief stays yours. The sources stay visible. The shape you asked for survives. That is the product, on an ordinary afternoon, not a promise about how the writing will score.
Who it is for
Draftwright is for an editor, a founder, or a small content team that is tired of starting from a blank chat. You already know the kind of piece you need. You may have a voice document, a few paragraphs you liked, and a clear memory of a draft you rejected last month. What you do not have is a desk that holds those decisions and brings them back when you start the next brief.
You are a good fit if someone on the team is still the person who approves. The desk drafts. You decide. Approve when the wording is something you would publish or send. Dislike when it is not, and say what was wrong, so the reason is specific enough to be useful later. If nobody is going to read the draft, the memory has nothing honest to learn from.
It fits more than one kind of morning. A question you want answered in your own materials. A page that has to be written. A long note you want shortened. A paragraph you want rewritten in the voice you have already approved. An afternoon of Deep Research with the sources listed. A Website Study of a single public site, when the new copy should sound compatible with that site and stay on it.
It is a poor fit if you want a detector to certify the writing, or a desk that publishes without you. A tone check, when one is turned on, is a mark for you to read. It does not guarantee what a detector will say. Approval is the decision that stores an example. Until you make it, the draft is a draft.
The person who still edits
The useful picture is an editor with a brief, not a pipeline that replaces them. They know the audience. They know which claims the company is allowed to make. They can feel when a paragraph has wandered off the assignment. Draftwright gives them a first version that respects the shape they asked for, and a way to mark the result so next time starts closer to their taste.
A founder who writes their own site can use the same desk. The “editor” is whoever reads the draft before it goes anywhere. The title on the sign-in does not matter. The habit does: choose the workflow, choose the model, read the draft, approve or dislike.
Agencies with more than one brand can keep those voices apart. A memory saved for one brand does not have to appear in another. A house rule that really is a house rule can be saved where both brands should see it. The trace on a draft shows which guidance was used, so you can stop before you approve a piece that picked up the wrong voice.
The work it replaces
The brief lived in a chat box. The brand rules lived in a document that was not pasted the next time. You explained the voice again. You explained the audience again. A numbered article came back without its sections, or with labels you never asked for leading each block. You fixed it by hand and forgot to save the fix anywhere the next draft could see.
A disliked tone came back the following week because nothing had stored the decision. The paragraph you loved was trapped in a thread you could not find. Research meant a pile of tabs, and the draft that followed did not say which tabs it had used. Studying a website meant copying pages into the chat and hoping the model stayed on that site instead of wandering.
Two editors made it worse. Each had a private chat. Each had a slightly different version of the rules. A client, or a founder, saw both versions in one week and asked which voice was real. There was no shared place where an approved example outranked a forgotten instruction.
Draftwright replaces that scatter with a desk. The workflow is a choice you can name. The model is a choice you can see. The shape of the brief is checked. The approval is stored. The sources for a research draft are listed. You still edit. You stop re-teaching the same lesson from scratch every Monday.
Six ways to start
Ask is for a question. You want an answer grounded in the guidance you have saved, not a fresh essay that ignores it. Create content is for a piece meant to be published or sent: a page, a post, a letter, a longer article. You say what it is, who it is for, and the shape it should take. Summarize takes material you provide and shortens it. Rewrite takes material you provide and writes it again, in the direction you ask for.
Deep Research is the start you choose when the piece needs public pages behind it. The draft comes back with the sources it used. You can read them. You can throw out a claim that the source does not support. Those pages belong to this request. They do not become your brand rules just because the search found them. If a fact should last, you save it on purpose.
Website Study is the start you choose when one public site is the subject. You give it that site. It reads there, and it writes new copy in a voice compatible with what it saw. It stays on the site you named. It will not take a private or local address, and it does not follow links off to other sites. The pages it reads are not filed as memory by the crawl. You still approve, or not, before anything is kept as an example.
Picking the workflow is part of the brief. A summary is not a research project. A website study is not a tour of the whole web. When the desk knows which job you meant, the draft has a chance of matching the afternoon you actually have. You can run a different workflow tomorrow without losing today’s piece. Each one stays on the account, in the workspace where you made it.
Choosing the start on purpose
If the material is already on your desk, Summarize or Rewrite. Pasting a page you wrote and asking for a shorter version is a different job from asking the desk to go and find pages. Mixing them is how a draft grows claims you did not bring.
If you need to see the sources, choose Deep Research and read the list before you approve. A confident paragraph with no pages under it is a reason to stop, not a reason to hurry. The list is there so the editor can do that check in the same sitting.
If the job is “write something that could live beside this public site”, choose Website Study and give it that one address. If you wanted three competitors compared, that is a different brief, and a website study will not quietly become one. You would be asking it to leave the site, which it will not do.
The model you chose
The console lists the models that are configured for your desk. That can include a local model, or Gemini, or OpenAI, when those are set up. You select the one this draft should use. The first draft uses that selection. You can see it on the response, so a colleague can ask “which model wrote this?” and you can answer from the screen.
Choice matters because the team already has preferences. One model may be the one you trust for a long page. Another may be the one you want for a short rewrite. Draftwright does not hide the choice inside a single unnamed engine. The person who briefed the piece is the person who picked the model, or they can see that someone else picked it.
A later pass, if you use one, still has to keep the assignment. You should not lose the sections you asked for because a second step tried to improve the tone. The shape is the contract you made when you wrote the brief. The model can change the sentences. It does not get to change the job.
If a model you wanted is not in the list, it is not configured. You pick from the ones that are. That is a better surprise in the morning than a draft that claims a model you do not have. The desk you brief is the desk you can see.
A draft that keeps its shape
You said what the piece is. A numbered article should come back numbered. A set of sections you named should still be those sections. A label you did not ask for should not take the top of the piece and turn it into a different kind of document. A long draft should not be replaced by a short fragment that happens to sound cleaner.
That is the content contract, in the words an editor actually cares about. The desk checks that the draft still matches the assignment before a new version can replace it. Planning, a revision, and a tone pass all have to clear that bar. You spend your time on the sentences, not on rebuilding the outline the brief already settled.
Write the shape in the brief the way you would say it to a colleague. “Six numbered sections.” “A short introduction, then the offer, then the ask.” “Do not add a headline style I did not request.” The clearer the shape, the easier it is to see that the draft kept it. Vague briefs produce vague arguments about whether the shape survived. Specific briefs make the check obvious.
You will still edit inside the shape. A section can be true to the outline and wrong in the claim. The contract is not a substitute for reading. It is the reason the thing you are reading is still the piece you commissioned, rather than a different piece that arrived wearing the same title.
What you notice when the shape slips
You notice a missing number. You notice a new label, bold at the top, that you never wrote into the brief. You notice a paragraph where a full section used to be. Those are the moments the contract is for. The draft you are shown should already have been held to them.
If you are unsure, compare the draft to the brief before you compare it to your taste. Taste is the approval. Structure is the assignment. Doing both in one glance is how a pretty fragment gets published in place of the article you needed.
When a draft keeps the shape and still sounds wrong, dislike it and say why. “Too formal.” “Invented a customer.” “Left out the price.” A reason the desk can store is more useful next time than a shrug. The next brief can then avoid that miss without you retyping the whole rule.
Approval that teaches the next draft
Approve a final version when you would stand behind the wording. That approval stores an example. The next draft, in the same kind of work, can draw on it. You are teaching with a piece you already accepted, which is clearer than teaching with a policy nobody opens.
Dislike stores the other lesson. You give a reason, and that reason is guidance to avoid. The following week, the miss you named is something the desk can steer away from. A vague dislike teaches less than a specific one. “Not this tone, because it sounds like a brochure” will do more than “no”.
Not every draft should become memory. A test, a joke, a half-finished experiment: leave them unapproved. Generations stay on your account so you can reopen the workspace. Lasting guidance is the smaller set you chose to save. You can edit a saved record later, turn it off, or delete it when it should stop guiding anyone.
Pasted notes and uploaded files are context for the current request. A text file or a PDF you attach to this brief grounds this draft. They do not become brand memory unless you save a record. That split keeps a one-off source from silently ruling next month. If the file holds a fact you want kept, save the fact. If it was only fuel for today, leave it as fuel for today.
What is worth saving
Save a voice note you are willing to repeat. Save a style rule you are tired of restating. Save a forbidden claim, an audience insight, a product fact, an approved example, a rejected example. Those are the kinds of memory the desk can keep. A request can draw on the scope you choose, and on house rules saved for everyone, when that retrieval is on.
Do not save a draft you have not read. Approval is a decision, and the next piece will treat it as one. If you are unsure, edit first, then approve the wording you actually want copied in spirit. The example will influence what comes next, which is the reason to be deliberate.
Review the records the way you review a style guide. Once a month is enough for a small team. Turn off the example that belonged to a campaign you have retired. The desk should sound like the company you are now, not like every approval you ever clicked.
Research that shows its sources
Deep Research returns the public pages it used, with the draft. You read the draft and the list together. A sentence you cannot match to a source is a sentence you can cut before you approve. The pages are evidence for this request. They are not filed as brand memory by the search itself.
That is the difference between a research afternoon and a pile of tabs. The tabs used to close. The draft remained, and nobody could say where a number came from. Here the list travels with the response. You can hand both to a colleague. They can disagree with a source. They can ask for a rewrite that drops the claim. The desk is not asking you to trust a footnote you cannot see.
Website Study is the narrower version of the same honesty. One public site, the one you supplied. New copy, written from what that site says, in a compatible voice. Private addresses are refused. Links off the site are not followed. The crawl does not silently become memory. If you want a line from the study to guide later drafts, you save it after you have read it, the same way you save any other example.
Use research when you need the outside world. Use saved memory when you need the company’s own voice and facts. A single draft can be informed by both: your approved examples, and a source list you can check. Approving the result stores your decision about the wording. It does not store the whole web behind it.
A normal working day
Tuesday begins with a sign-in. Yesterday’s workspace is still there if the piece was unfinished. Today’s job is a new page. You choose Create content, and you choose the model you want for a first draft of this length. You write the brief the way you would brief a person: who it is for, what it must say, the six sections, and the claims you will not make. You paste a short note from the founder that applies only to this page.
The draft comes back in those sections. The trace shows the model and the saved guidance that was used. The founder’s paste stayed on this request. A voice example you approved last month is in the trace, which is what you wanted. You fix one over-eager sentence, then approve the final version. That approval is now an example the next page can learn from.
After lunch you run Deep Research for a question the page raised. The response includes the public pages. You drop a claim the nearest source does not support, and you do not approve the research draft as brand voice. You only needed the sources. The page you already approved stays the example. The research stays the research.
Thursday a colleague signs in as themselves. They cannot open your workspace. They start a Website Study of one public site the team is allowed to read, and they get copy that stays on that site. They dislike the first version because it invented a customer story, and they say so. That reason is stored as something to avoid. Their history is theirs. Your Tuesday page is not in it.
Friday you start a related page. The model you choose is the same. The trace shows Tuesday’s approved example and the guidance to avoid the invented customer. The sections you asked for are intact. You edit less than you edited on Tuesday. That is the day the desk is for: not a perfect first sentence, a second piece that remembers the decision you already made.
An afternoon on one site
Website Study is a good Thursday. You have a public site and a need for new copy that could sit beside it: a summary, a page in a compatible voice, a clearer version of an offer the site already makes. You give the address. You read what comes back. You check that it did not stroll off into other domains, because it should not have.
You treat the result as a draft. If a sentence overclaims, you cut it before anyone approves. If the voice is right and the facts are fair, you approve the lines you mean to keep. The crawl itself has not written those lines into memory. Your approval has.
If the address is private or local, the study refuses it. That is the correct stop. The desk is for public sites you can name, not for anything sitting behind a login or on a machine in the office.
What stays in your control
You choose the workflow and the model. You write the brief, including the shape. You decide what to paste or upload for this request only, and what to save as memory that should last. You approve, or you dislike, or you do neither and leave the draft as a draft on your account.
You choose the scope. One brand’s voice, facts, and examples stay with that brand. House rules are the ones you deliberately saved for more than one brand. The trace shows what was used, so you can see a mistake before you approve it. Another person’s sign-in does not see your workspaces. A shared memory appears because it was shared, not because the accounts are the same.
You choose what research is allowed to become. Sources come back with the draft. They enter memory only if you save a record, or if an admin proposes candidates from a public site and a person reviews them before they guide anyone. A crawl is not a silent edit of the style guide.
You remain the editor. The desk will not publish for you. It will not decide that a tone mark is permission to ship. If a tone check is on, you read the mark and you still approve, or not. The next draft learns from that human decision, which is the only teaching this product claims.
Two people, two brands
Give each editor their own sign-in. Give each brand its own scope. Generate in the first, approve an example, then generate in the second and read the trace. The first approval should not appear in the second. Generate with the house rules and see those, because you saved them as house rules.
That hour is how you learn the desk is safe to share across a small agency. The danger was never that two people typed at once. The danger was a single chat where every brand’s instructions were pasted on top of each other. Scopes are the way out of that pile.
An editor who picks the wrong scope will see the wrong memories in the trace. They can stop before they approve. The trace is the check. Teach people to read it the first week, and the wrong-brand draft gets caught while it is still a draft.
Built with
Draftwright is a Next.js desk in TypeScript, with Python doing the writing work and PostgreSQL holding accounts, drafts, and memory. The model you selected is the one the first draft uses, from the providers configured on the desk. You do not operate that stack to write a paragraph. You sign in, choose the workflow, and read.
The pieces a technical partner would ask about are those four, plus the record of an approval and the source list on a research draft. They exist so a decision you made on Tuesday can still be found on Friday, and so a Deep Research or Website Study can show its work. The editing stays with you.
A product in this shape, from TechMamba
Draftwright is the kind of product TechMamba builds. An editor can explain it without a diagram: six ways to start, the model I chose, a draft that keeps its shape, an approval the next draft can learn from, and research that shows its sources. The desk is there so the human decision sticks.
If you want a writing desk in this shape, bring a brief you already know how to edit and an example of wording you would approve. Tell us who signs in, whether more than one brand has to stay apart, and which of the six starts you actually need in the first month. We will map the screen around that: the workflow, the model, the draft, the trace, the approval, and the sources.
Run it on a piece you can judge. If the second draft is closer to the decision you recorded, and you can see why, the brief is the right shape. Bring us that piece. We will build the desk that remembers what you approved.
Questions worth asking before you brief it
Who approves? Which workflows do you need first? Which models should be on the console? What kinds of memory are you willing to save, and who may turn a record off? Do brands need separate scopes? Should research be allowed to enter memory only after a person reviews it? Those questions decide the desk more than any setting hidden underneath.
Then run the pass. Sign in. Choose Create content and a model. Brief a numbered piece. Read the draft against the brief before you read it for taste. Approve one version you mean. Dislike one version with a real reason. Start a second brief and see whether the trace shows those decisions. Run Deep Research and read the sources. Run Website Study on one public site and confirm it stayed there.
Bring a bad example on purpose. A brief that asks for sections. A paste that should not become permanent memory. A claim you want forbidden. The desk is working if those behave differently from each other, in ways you can point at on the screen. If everything is dumped into the same pot, you are back in the chat box.
TechMamba can take the brief from the editor’s side of the table. You do not have to arrive with a pipeline diagram. Arrive with the piece, the voice you will stand behind, and the mistakes you do not want repeated. We will say which parts of Draftwright already match that week, and what you should decide before a second brand shares the desk.
How a team settles in
The first week is one editor, one workflow, and one model. They bring a piece they have written before, so they can tell a good draft from a fluent wrong one. They approve sparingly. They write dislike reasons in full sentences. They look at the trace every time, until reading it is a habit rather than a feature tour.
The second week adds a second workflow. Often that is Deep Research or Rewrite, depending on whether the team is short of sources or short of voice. They learn that pasted material stays on the request, and that saving a record is a separate, deliberate act. Someone tries to upload a file and then checks that it did not become a permanent rule. That check is the lesson.
A second person joins with their own sign-in. If there is a second brand, it gets its own scope before that person approves anything. You prove the split once, in an hour, and then you trust it. House rules are written down as house rules, not pasted into whichever brief is open.
After a month the style guide is partly the records themselves: approved examples, rejected ones, facts, forbidden claims. Someone still owns the review. The desk does not retire the editor. It gives the editor a place where Tuesday’s decision is still available when Friday’s brief begins.
What you still do by hand
You still cut the sentence that overreaches. You still check a source before you let a number stand. You still decide that this draft is the one the company will say out loud. The desk can hold the shape, retrieve the guidance, and list the pages. It cannot want the right thing for the reader. That remains the job.
You still write the brief. A careless brief, approved in a hurry, becomes an example you will see again. The settle-in period is when the team learns to approve slowly. Speed comes after the records are ones you like.
You still talk to each other about voice. The trace helps the conversation, because you can point at the guidance that was used. It does not replace the conversation. A record that surprises both editors should be turned off until you agree what it was for.
A second piece, closer to the decision you recorded
The test of the desk is not the first draft. The first draft is where you find out whether the shape held and whether the model you chose is one you can edit. The test is the next brief, in the same kind of work, after you have approved one piece and disliked another. You should see those decisions in the trace. You should spend less time restating the voice, and more time on the sentences that are actually new.
Pick a second piece you already know how to judge. Ask for the same kind of sections. Read the trace before you read for polish. If Tuesday’s approved example is there, and the invented customer you disliked is being avoided, the memory is doing the job you saved it for. If the trace is full of guidance you do not recognise, turn those records off before you approve anything else. A noisy memory is worse than a short one.
Keep research in its lane on that second piece. If you need sources, run Deep Research and read them. Do not approve the research draft as if it were the brand voice, unless you have edited it into wording you would sign. Website Study, if you use it the same week, should still be one site. A busy week is when people are tempted to treat every workflow as the same button. The six starts stay different so the second piece does not inherit a crawl you never meant to keep.
Show the second draft to the person who will publish it, with the brief beside it. The conversation gets shorter when you can say which model ran, which example was used, and which sources you checked. That is the settle-in you are aiming at. Not a desk that writes the final line, a desk that makes the editorial conversation about the piece instead of about where the rules were saved.
If the second piece is not closer, look at what you approved. A hurried approval teaches the hurry. A dislike with no reason teaches nothing you can point at. Fix the records, then run a third brief before you decide the desk is wrong for the team. The product is the loop: brief, draft, your mark, the next brief. Give the loop one honest cycle before you judge it.
What good looks like on the screen
Good looks like a workflow you meant, a model you recognise, and a draft whose sections match the brief. The trace names the memories that were used. Sources, if this was Deep Research, are listed. A Website Study names the one site and does not pretend to have read the rest of the web. An approval is rare enough that it means something.
Good looks like a second draft that avoids the mistake you disliked, without losing the shape. It looks like a pasted note that helped today and did not become law. It looks like a colleague’s workspace staying theirs. It looks like a house rule appearing only where you saved it as a house rule.
You can explain a draft to someone who was not in the room. Here is the brief. Here is the model. Here is what the memory contributed. Here are the sources. Here is what I approved. That explanation is the product working. A draft you cannot explain is a draft you should not approve, however smooth it sounds.
Over a month, good looks like less re-teaching and more editing. The voice document is no longer a file you forget to paste. The decisions are on the desk. You still read every piece you are willing to sign. Draftwright’s job is to make the next one start from that signature, and to show its work while you decide.