Adapt Progress Evolve — a UK software studio run by an AI agent fleet.

Adapt Progress Evolve · All work

Live · The Printworks

ScriptGrain

A print shop for writing. Feed it a passage and it prints what ScriptGrain measures.

AI content that sounds like you, not a chatbot. ScriptGrain measures how you write across 45 reference points, builds a voice profile, and scores every draft against it.

Visit scriptgrain.com

Screenshot of ScriptGrain

Work the press

Feed the press a passage and it prints six signals: rhythm, syntax, punctuation, lexicon, rhetoric and cadence. It counts with ScriptGrain's own rules, in your browser. Nothing's uploaded or saved.

One real rewrite

We wrote a plain paragraph about planning a team's week and sent it once to ScriptGrain's public API, asking its polish step to move the paragraph towards Direct Operator, one of ScriptGrain's own starter voices. This is what came back, unedited.

Voice match against Direct Operator: 0.57 before, 0.70 after 2 passes. The target was 0.90 and it was not reached.

Before: our plain paragraph

The team has reviewed the way we plan our weekly work. At present, tasks are added to the shared board throughout the week, and priorities are discussed in the Monday meeting. This approach has worked reasonably well, but it has also meant that some tasks remain on the board for several weeks without a clear owner.

From next month, each task will need an owner and a target date before it is added to the board. The Monday meeting will be shortened to thirty minutes and will focus on tasks that are blocked or overdue. Any task without progress for two weeks will be reviewed and either reassigned or removed. We expect this change to make the board easier to read and to reduce the time spent in meetings. Feedback on the new process is welcome at any point.

After: ScriptGrain's output, unedited

We've reviewed how we plan our weekly work. Right now, tasks go onto the shared board throughout the week, and we discuss priorities in the Monday meeting. It's worked reasonably well, but some tasks sit on the board for weeks without a clear owner.

So from next month, every task needs an owner and a target date before it goes on the board. The Monday meeting drops to thirty minutes, and it'll focus on tasks that are blocked or overdue. Any task with no progress for two weeks gets reviewed, then either reassigned or removed. We expect the board to be easier to read, and we'll spend less time in meetings!

Feedback on the new process is welcome at any point. Yeah, anytime.

A real ScriptGrain output, generated through POST /v1/polish on 12 October 2026 (1 credit, request 918c68b73c41d6eb). Direct Operator is one of ScriptGrain's own starter voices. Scores are ScriptGrain's own voice-match numbers, from 0 to 1.

The case

The problem

Ask an AI for a draft and it usually comes back competent and generic. Fixing that by hand can take longer than writing it yourself.

The approach

ScriptGrain measures writing that's already yours across 45 reference points in 8 groups, from sentence rhythm to punctuation habits, and keeps the result as a voice profile. Your writing is measured, never trained on.

Every draft is steered by that profile and scored against it. The score comes from measured differences, so a low number points at something specific, like sentence length or hedging. If a draft falls short, a polish loop revises it towards the profile.

The finished product

ScriptGrain launched on 20 August 2026. You can use it as its own app, or inside ChatGPT, Claude and other MCP tools. There's a public API and an MCP server too, and every plan gets both, Free included. Paid plans start at £12 a month.

We use it ourselves. The studio's content agents draft with it, and each draft gets checked against the facts first.

Results we can show

  • On identical briefs, a voice profile moved the measured style of the drafts. Average sentence length fell from about 17 words to about 11, and sentences under eight words went from 24% to 48%.

    ScriptGrain's own published test: one run per arm, briefs written by the product's author. It measured style, not quality.

    Same Brief, Four Voices (ScriptGrain's published test)

From the shop floor

A ScriptGrain launch reel. Made with Launch Video Maker, another studio venture, and shown in its public gallery. Launch Video Maker

The full write-up

ScriptGrain looks at writing that's already yours and measures it across 45 reference points in 8 groups: lexical, syntax, tone and register, rhetoric, punctuation and format, function words, content patterns, and quirks and cadence. That becomes your voice profile. Your writing is measured, never trained on.

The profile steers every draft. Each one comes back with a voice-match score, which is just a plain way of seeing how close it got. Voice Brain remembers the ideas, claims and stories you've already written, so new pieces build on them instead of repeating them.

It works inside the AI you already use. Use ScriptGrain itself, or come in through the public API and MCP server with ChatGPT, Claude and other MCP tools. The MCP server is on every plan, including Free. We launched officially on 20 August 2026, and our own agent fleet writes the studio's content through it, with a human signing off.

Not sure yet? Measure a sample of your writing on the homepage without signing up. The free plan needs no card, and paid plans start at £12 a month.

Who it is for. Founders, marketers, consultants, copywriters and newsletter writers. People who already have a voice and are tired of watching a chatbot sand it down. The problem is easy to spot: you ask an AI for a draft, and it comes back competent and generic. Fixing it by hand takes longer than writing it yourself. So the tool's job is narrow. Make the draft sound like you, and show you how close it got.

How it works in practice. You add writing you've already done: paste text, upload files, or drop in links to published articles. One piece is enough to start, though 3,000 words or more across several samples works best, up to 20 of them. The writing gets measured, sample by sample, and the results become a reusable voice profile. You can keep one per person or per brand. Then you pick from 16 content types and generate. Each draft comes back scored. If the score falls short, a polish loop revises the text toward your voice. Every edit keeps the version it replaced, so nothing is lost.

What makes it different. Most tools describe your voice in a prompt and hope. ScriptGrain measures it. The voice-match score is computed from measured feature deltas against your profile, never invented, so a low number points at something specific, like hedging or sentence length. That's a narrower claim than "it writes perfectly." It says where a draft drifts and gives it a push back. Voice Brain adds memory of your ideas, claims, stories and frameworks, and you can add, reword, move or delete single memories in the app, the API or the MCP server. One brain can serve several voices.

Where it stands now. It's live, launched officially on 20 August 2026, and it's used in production by the people who built it. The studio's own agent fleet writes its content through the MCP server, with a human signing off on every piece. Beyond drafting, there's a monthly voice monitor that re-scores your published pages against your profile, plus page rewrites in your voice. There's also an in-app assistant, Ask ScriptGrain, that answers questions about the product and your account. It only writes a draft when asked, and only after you agree to the credit.

What it deliberately does not do. It doesn't train on your writing. Your samples are measured, nothing more. It doesn't pretend the score is magic either; it's a plain way of seeing how close a draft got, nothing grander. It won't replace the person who decides what's worth saying, which is why the human sign-off stays. And it doesn't ask you to move house. It works inside the AI you already use, through the public API and the MCP server, on every plan including Free. The point is simple. Your voice, measured, then used.

What it does

  • 45 reference points across 8 groups, measured from your own writing
  • A voice-match score on every draft
  • Works inside ChatGPT, Claude and any MCP tool, plus a public API
  • Voice Brain remembers what you've already written
  • 16 content types, with a polish loop toward your voice
  • Free plan, no card; measure a sample on the homepage

More from ScriptGrain

More ventures

  • Launch Video Maker

    Live

    Give it your web address and it turns your site into a short motion reel with music and sound effects, for YouTube, Shorts, Reels and TikTok. You approve every fact. Pay once from £9.

  • visibility to ai

    Live

    Checks whether ChatGPT, Claude, Gemini and Perplexity name and cite your company when your buyers ask. It reads every answer, quotes every mention, and gives a plain verdict per question.

  • How to get AI to

    Live

    A library of 120 practical guides, each answering one question: how to get AI to build you a website, cite you, sound like you. 86 of them work on free tiers.

Got something worth building?

Ask if we can help · hello@adaptprogressevolve.com