open source · MIT licensed

replicateme

CLI that learns your writing style and generates messages that sound like you.
From formal emails to casual texts.

Terminal
$ brew install jschell12/tap/replicateme
View on GitHub Get started Something broken? Let me know

How it works

1

Ingest

Point it at your messages. 9 sources: iMessage, Slack, Gmail, Twitter, Discord, Reddit, Instagram, GitHub, TikTok.

$ replicateme ingest --source imessage
Scanning ~/Library/Messages/chat.db... Found 12,847 sent messages Analyzing writing patterns... Style profile saved to ~/.replicateme/profile.json
2

Analyze

It builds a fingerprint of how you actually write: capitalization habits, punctuation use, common shortcuts, typical errors.

$ replicateme profile
Style Fingerprint ----------------- Avg message length: 31 chars Capitalizes first word: 81% of the time Ends with period: 1% of the time Uses question marks: 17% of the time Uses emoji: 2% of the time Short fragment messages: 58% of the time Lowercase "i" instead of "I": 27% of the time Writing habits ---------- "im" without apostrophe 843 occurrences "dont" without apostrophe 312 occurrences "thats" without apostrophe 198 occurrences
3

Generate

Give it a prompt and it writes back in your voice, using your style profile to match your patterns.

$ replicateme gen "Friend asks: want to hit up that new brewery this weekend?"
Generated 3 variants (tone: casual) 1. Yeah im down for that. What day were you thinking? 2. yeah could be down for that. what day were you thinking? 3. yeah im down, what day were you thinking

Tone control

Slide between casual and formal. Both sound like you.

Casual 50% Formal
0%

"I'm going to be about 15 minutes late. Can you grab me a coffee?"

Formal: your voice, clean grammar
25%

"Im going to be about 15 minutes late. Can you grab me a coffee?"

Work email: mostly clean, a few habits slip through
50%

"im gonna be like 15 min late. can you grab me a coffee"

Slack message: natural mix of your style
75%

"im gonna be like 15 min late can you grab me a coffee"

Group chat: dropped punctuation, fragments
100%

"im gonna be like 15 min late can u grab me a coffee"

Text to a friend: full casual, all of it

Features

🔒

Privacy-first

All data stays on your machine. The CLI has no accounts, no telemetry, no tracking. This site keeps an anonymous page view count, nothing else.

🔍

Style fingerprinting

Analyzes your real patterns: capitalization, punctuation, contractions, phrases, typical errors. Not a generic prompt.

🎛

Tone control

Slide between casual and formal. Plus granular writing style toggles: misspellings, grammar shortcuts, missing apostrophes, lowercase i, dropped punctuation, and fragments.

💬

Multi-platform

Generate for iMessage, Slack, email, GitHub commits, Twitter, TikTok. Each platform gets appropriate tone and length.

📥

9 data sources

iMessage, Slack, Gmail, Twitter, Discord, Reddit, Instagram, GitHub, and TikTok. Download your data archive, feed it to the CLI.

🔑

4 LLM providers

Anthropic API, OpenAI API, Claude CLI (use your Max subscription, no API cost), or Ollama for fully local inference. Any OpenAI-compatible endpoint works too.

👤

Persona specs

Load a markdown persona file with --persona to describe how you write in prose. Goes beyond statistics into voice and tone.

📋

Git hooks + clipboard

Auto-generate commit messages in your style with the included git hook. Copy any variant to clipboard with --copy.

📦

Works anywhere

Export a rule file and drop it into Claude Code, Cursor, or ChatGPT on any laptop. If the tool is installed there, enrich the rule with work data. No raw messages ever leave your machine.

Your data stays yours

✓

Messages are read from your local databases. Nothing is uploaded.

✓

Style profiles stored at ~/.replicateme/ on your machine.

✓

The only network call is to your LLM provider with the style profile for generation. No raw messages sent.

✓

The CLI: no telemetry, no analytics, no accounts, no tracking. This website keeps an anonymous, cookieless count of page views and install clicks. No personal data, no cookies, no third parties.

✓

MIT licensed. Single binary, no dependencies. Fork it, audit it, run it airgapped if you want.

Your machine
iMessage DB
↓
replicateme CLI
↓
~/.replicateme/profile.json
style profile only
→
LLM API
Anthropic or OpenAI
↓
Generated text

Data sources

iMessage live Reads local ~/Library/Messages/chat.db
Slack live Workspace export (ZIP)
Gmail live Google Takeout (mbox)
Twitter/X live Data archive (tweets + DMs)
Discord live Data export (ZIP)
Reddit live Data archive (posts + comments)
Instagram live Data download (DMs, captions, comments)
GitHub live Local git repos (commit messages)
TikTok live Data download (comments + DMs)

Get started

Terminal
# install via homebrew or go
$ brew install jschell12/tap/replicateme
# or: go install github.com/jschell12/replicateme/cmd/replicateme@latest
 
# set your API key
$ export ANTHROPIC_API_KEY=sk-ant-...
 
# ingest your messages
$ replicateme ingest --source imessage
 
# generate and copy to clipboard
$ replicateme gen --copy "Reply to: hey are you coming tonight?"
Generated 3 variants (tone: casual) 1. yeah ill be there 2. yeah im coming 3. yep on my way soon Copied variant 1 to clipboard
 
# use a persona spec for even better results
$ replicateme gen --persona ~/persona.md "Tell the team the deploy is done"
 
# auto-generate commit messages with the git hook
$ cp examples/prepare-commit-msg .git/hooks/
View source on GitHub Report a bug