mimicry
Rewrites AI drafts in your own voice, with Jev as the judge in a bounded loop that checks style and that meaning stays intact.
Overview
Mimicry turns a rough opinion into a clear draft and then matches it to your own past writing, with less empty framing and repeated explanation. The source draft supplies meaning and your reference samples supply style only: facts, requests, stance, certainty and commitments must stay intact, and anecdotes from references must not become invented personal experiences. It runs as a local web workspace or a Chrome extension that adds a button inside X's post, reply and quote boxes.
An OpenAI model (gpt-5.6-luna by default) writes rewrites, and Jev (jev-1.13.0 by default) acts as the style judge in a bounded feedback loop, rating attempts and checking meaning, so only rewrites that keep the meaning are kept. The result separates the AI rewrite from 'in your voice', and an expandable view shows the kept version's diff against the original and how the loop decided, with all attempts, ratings and meaning checks visible. Keys stay server-side; references and drafts are sent to OpenAI and TypeSafe, and runs are saved locally.
What the decision model decides
| Type | Question | Answers |
|---|---|---|
| Score | How closely does this rewrite match the user's voice?(inferred) | — |
| Yes / no | Does this rewrite preserve the original meaning, facts and commitments?(inferred) | — |
How it works
- Add your writing samples and a draft opinion
- LLM writes candidate rewrites
- Jev rates style and checks meaning for each attempt
- Loop stops within a bound; best meaning-preserving version is kept
- Diff and decision trail are shown
Strengths
- Meaning checks guard against drift
- Transparent attempts and ratings
- Works inside X via a Chrome extension
Caveats
- Requires both OpenAI and TypeSafe keys
- Text is sent to both providers
- Style judgments are model opinions
Getting started
uv sync --extra dev, create .env with OPENAI_API_KEY and TYPESAFE_API_KEY, then uv run mimicry-web and open http://127.0.0.1:2719.At a glance
- Type
- Project
- Status
- prototype
- Language
- Python
- Runtime
- local web app and Chrome extension
- Has code
- yes
Use cases and integrations
How curated lists describe it
Rewrite AI drafts in your own voice with a bounded TypeSafe feedback loop.
hellogumbo/awesome-jev
Frequently asked questions
What does Jev do here?
It judges rewrites in a bounded loop, rating style and checking that meaning is preserved.
Who writes the text?
An OpenAI model; Jev only judges.
Where does it work?
In a local workspace or inside X's compose boxes via the Chrome extension.
Related projects
Editorial QA tool where Jev judges per fact whether an edit preserved the original, with a confidence gate and human review fallback.
ProjectbetaJavaProse linter for AI writing tells: regex handles 18, Jev judges 15 as Nouls in one call, shipped as a Claude Code Stop hook.
HookbetaPythonPublished Jev client and claim check from the Fake/Real evidence-review site, where Jev filters sources and judges each claim.
ProjectproductionTypeScriptSRT subtitle translator that translates with any OpenRouter LLM and asks Jev a review Noul per line to flag defects before export.
ProjectbetaPythonEvidence-first experiment-memory skill for coding agents, with an optional Jev audit layer that checks provenance and claim support.
Agent skillbetaPythonjevctl: Node CLI for typed Jev judgments (verify claims, screen agent hijacks, classify) with exit codes for CI gating.
CLIbetaTypeScript