Gentle-AI
Gentleman-Programming/gentle-aiReviewed by AI
Checked by RepoPicks Radar on
What is Gentle-AI?
Gentle-AI writes the configuration for AI coding assistants you already have, such as Claude Code, Cursor, Codex, Copilot or Gemini CLI. It adds a memory store so the assistant keeps project decisions between sessions, a step-by-step working method, a library of skills, MCP servers, personas and a review pass over finished work that is on by default. It does not install any assistant for you and it snapshots each configuration file before writing. For developers: install with Homebrew or a shell script on Mac and Linux, or with Go on Windows, then run `gentle-ai`.
How easy is it to install Gentle-AI?
B · installer plus one simple step · macOS
- Windows C
- macOS B
- Linux B
On Mac the ready binary comes from `brew install gentleman-programming/tap/gentle-ai` (Linux uses a shell script), after which you run `gentle-ai` once to pick your agents and components — so it is an install plus one step, in the terminal. Windows has no binary in the releases: the README installs it with `go install`, which means putting Go 1.25.10 or later on the machine and fixing your PATH by hand.
Gentle-AI on GitHub
- Stars
- 8,741on Nov 10, 2026
- Downloads
- 552,4096,702 this week
- Growth
- +173stars in 3 days
Stars over time
Facts
- Runs on
- macOS, Linux, Windows
- You need to install
- Nothing extra
- To build from source
- Node.js, Go
- Disk space
- 18 MB–53 MB · from the installer
- Accounts or keys
- None
- Cost
- Free, API costs extra · The tool itself is free (MIT), but it only does something on top of an AI coding assistant, and the usual ones are paid subscriptions or paid API usage.
- License
- MIT (Permissive)
- Last change on GitHub
- Nov 10, 2026
Gentle-AI risk signals (7)
- Attentioninstall pipes script from internet to shell
README.md, Get started: `curl -fsSL https://raw.githubusercontent.com/Gentleman-Programming/gentle-ai/main/scripts/install.sh | bash`. The script runs whatever is on the main branch at that moment, and it also edits your shell startup file ("install.sh adds one marked line (# gentle-ai installer) to the startup file your shell reads: ~/.zshrc for zsh..."); `install.ps1` "adds the directory to your user Path".
AI review
- Attentiontelemetry on by default
README.md links a documentation page called "Telemetry — What we count, and how to turn it off" (docs/telemetry.md). Counting happens unless you turn it off, and what is counted is not described in the README itself.
AI review
- Attentionrewrites your agent configuration
The repository description says it "configures the AI coding agents you already use: Claude Code, Cursor, OpenCode, Codex, Pi, and more", and the README says "Your configs are snapshotted before every write". It edits files that decide how your coding assistant behaves, including MCP servers and skills; the listed guards are a "Security deny-list" that "Blocks ~/.ssh, .env and credential files" and the config backups.
AI review
- Infodependency code committed to the repository
The top-level file list includes `node_modules/` inside a Go project whose package.json has no dependencies and only a failing test script ("test": "echo \"Error: no test specified\" && exit 1"). Third-party JavaScript files are therefore stored in the repository itself.
AI review
- Infoself reported single run benchmarks
README.md, "Measured": the before/after table is from the author's own runs and states "each cell is a single run (n=1), and cost is API-equivalent". The numbers also come from a different project, Gentle Shell, not from this binary alone.
AI review
- Inforeview mode enabled by default
README.md: "Receipt-Driven Development (RDD) is on by default and opt-out: run `gentle-ai review mode disable` to turn it off." Extra review steps mean extra model calls, and so extra spending on your agent, unless you disable it.
AI review
- Infotrademark limits on top of mit
README.md trademark notice: "The Gentle AI™ and Engram™ names and logos are trademarks of Alan Buscaglia... The MIT License applies to the code; it does not permit implying endorsement or official affiliation." The code is MIT, the names are not.
AI review
Evidence is shown in its original language.
What Radar could not check about Gentle-AI
- What the telemetry sends and whether turning it off really stops it
- Which of the 17 advertised assistants work fully, since the README says features "can differ" per integration
- How much extra the default review and memory steps add to your agent's bill
- Whether the configuration backups let you get your previous setup back cleanly
- Whether the Windows path works without Go, given there is no Windows binary in the latest releases