About Deft
Better AI Writing
Deft is a writing system built around Distribution Fine-Tuning. It turns rough source material into cleaner prose. We currently offer a simple text generator and a text rewriter as our first public tools; we also train custom models for enterprises that need custom workflows or custom styles.
System
What makes it different
Most writing products expose a generic model and ask the user to become better at prompting. Deft takes a different route: a custom model, built on an innovative training algorithm, trained to reproduce the shape and texture of strong finished writing.
That algorithm is Distribution Fine-Tuning. Ordinary fine-tuning pushes a model toward the single most agreeable answer, which is why so much AI prose arrives in the same cadence and the same hedged phrasing. DFT fits the model to the distribution of real human writing rather than its average, so what comes back varies the way real writing varies.
- 01
Start with rough material
Paste a draft, outline, fragment, or generic model output into the editor. Deft is designed for unfinished material that needs a stronger final shape.
- 02
Generate through DFT
The system routes the request through a fine-tuned writing model that has learned the distribution of finished, human-readable prose rather than relying only on a longer prompt.
- 03
Edit the result
The console returns an editable draft with lightweight metrics and optional checking tools, so the writer can keep judgment and revision at the center of the workflow.
FAQ
Frequently asked questions
What is Deft?
Deft is a writing tool built on distribution fine-tuning. You give it a prompt or a draft and it returns a finished piece of writing, produced by a model trained to match the range of real human prose rather than the averaged register most language models default to. We currently offer a simple text generator and a text rewriter as our first public tools. We also train custom models for enterprises that need custom workflows or custom styles.
What is distribution fine-tuning?
Instruction-tuned models are optimized to return the single most agreeable answer, which is why their prose converges on the same cadence and the same hedged phrasing. Distribution fine-tuning trains on the spread of human writing instead of its average, so the output varies the way real writing varies.
How is Deft different from ChatGPT or Claude?
General assistants are very good at reasoning and very bad at writing. Deft is narrower on purpose. It plans an outline first, then writes section by section, which is why it holds its shape across thousands of words instead of drifting into summary.
Does writing from Deft sound human?
That is the whole design goal. We ran a sample of 100 model outputs through a leading AI detector and they scored as 100 percent human-written. But the honest test is reading it, so we published the raw outputs side by side with GPT-5.5 and Claude Opus 4.8.
Is Deft an AI humanizer or a detector bypass?
Not in the usual sense. Humanizers take finished AI text and swap synonyms until a classifier changes its mind, and the writing is just as bad afterward. Deft generates differently in the first place, so there is no evasion step because there is nothing to evade. The result is that DFT is the least detectable AI model on the market, but that follows from writing well rather than from a trick applied at the end.
Can I try Deft without paying?
Yes. You get three free generations before signing in, and the writing console is free to use after you make an account. Usage-based billing applies to the server-to-server API.
How much does the Deft API cost?
API generation is billed by usage: $2.50 per million preprocessing input tokens and $12 per million final DFT output and preprocessing thinking tokens, rounded up to the nearest cent with a one-cent minimum per request. Judge and post-processing model tokens are not charged.
Am I charged for failed generations?
No. A failed generation releases its reservation and is never billed.
Do I need approval for an API key?
Yes. Request access from the API tab in your account with the site you are building and what you intend to use it for. Once approved you can create keys yourself, buy credits, and turn on auto-recharge. A raw key is shown once at creation and never again.