Sloth-Butler: Replies to the client messages you dread
Description and use
Sloth‑Butler is a web app that uses a fine‑tuned Qwen3‑8B LoRA model to generate polite yet firm client replies, enforce user‑defined terms, schedule follow‑ups, and provide voice notes.
Problem
Freelancers and service sellers struggle to craft polite yet firm replies to client requests for free work, discounts, late invoices, or other unfavorable terms, leading to delayed responses and potential revenue loss.
Outcome
The fine‑tuned model increased the percentage of replies that pass all five evaluation criteria from 41% to 63%, improved sendable rate from 84% to 93%, and reduced average reply length from 51 to 34 words.
Workflow
- Open https://sloth-butler.vercel.app
- Press "Write a reply"
- Fill and save "Your terms" once
- Select situation and paste client message
- Add amount/deadline, tone, language
- Press "Write my reply"
- Copy, play or download mp3
- For late invoices, press "Plan follow‑ups" and use board
- After sending, log hours and money protected
Evidence
Base Qwen3-8B and the fine‑tune answered the same 100 held‑out messages with the same prompt, temperature 0.7 and seed 2026. A judge model (gpt-oss-120b) scored each reply 1 to 5 per criterion; a pass is 4 or more.
The model can still get numbers wrong. For example, it sometimes offers a discount when the job is already at your minimum price. The server‑side guard catches that and corrects the reply, but it's a code check, not the model getting it right.
Six languages: English, Spanish, French, Portuguese, German and Nigerian Pidgin.
Select a tone (soft, firm, maybe‑later) and a language.
The training loop is adapted from tinker‑cookbook's supervised recipe.
- Record date: 2026-10-25T14:11:10.650Z