JotBot - AI-powered code documentation

JotBot seamlessly auto-generates code documentation for your Go and TypeScript projects, bridging the gap between comprehensive code and insightful documentation.

The documentation of this repository was entirely generated by JotBot using gpt-4 but currently (2023/07/18) I get the best results using gpt-3.5-turbo.

OpenAI's gpt-3.5 and gpt-4 models demonstrate fluctuations in quality over time. Determining which model excels at any given instance can be challenging, but gpt-4 often yields more consistent results.

For general applications, I recommend using gpt-3.5-turbo-16k as a default, and temporarliy switching to gpt-4 in scenarios where gpt-3.5 might fall short.

gpt-4-1106-preview is the best model for generating documentation so far. The default model of JotBot is gpt-3.5-turbo which is gives the best value for money.

You can find generated TypeScript documentation in packages/jotbot.

Quick Start

Install

go install github.com/modernice/jotbot/cmd/jotbot@latest

Add TypeScript Support

npm i -g jotbot-ts@latest
pnpm i -g jotbot-ts@latest

Use

Within your Go and/or TypeScript codebase, run:

jotbot generate --key OPENAI_API_KEY

---

Features

• Generate documentation for Go and TypeScript codebases
• Customize glob patterns for included and excluded files
• Filter code symbols by matching regular expressions
• Limit the number of files to generate documentation for
• Run in dry mode to preview changes without applying them
• Control the AI model and token limits used for generating documentation
• Optionally commit changes to a Git branch

Models

JotBot utilizes OpenAI's GPT models to generate documentation. By default, JotBot uses the gpt-3.5-turbo model which provides good results at a cost-effective rate. However, the best results are produced by gpt-4 and text-davinci-003, which are 10-30x more expensive than gpt-3.5-turbo.

You are free to choose any model from the OpenAI Docs and pass it via the --model|-m flag:

jotbot generate -m text-davinci-003

Installation

Via go install

If you have Go installed, you can simply install JotBot using go install:

go install github.com/modernice/jotbot/cmd/jotbot@latest

Standalone Binary

You can download the latest binary from the Releases page.

TypeScript Support

To enable TypeScript (and JavaScript) support, you also need to install the jotbot-ts npm package.

npm install -g jotbot-ts
pnpm install -g jotbot-ts

Usage

To generate missing documentation for your codebase, run the following command:

jotbot generate [options]

By default, this command will find all Go and TypeScript (and JavaScript) files in the current and nested directories and generate documentation for them. Excluded from the search are by default:

• **/.*/**
• **/dist/**
• **/node_modules/**
• **/vendor/**
• **/testdata/**
• **/test/**
• **/tests/**
• **/*.pb.go

To-Do

• Configurable OpenAI settings (temperature, top_p etc.)
• Any ideas? open an issue or start a discussion

CLI options

jotbot --help

Option	Description	Default
--root	Root directory of the repository	"."
--include, -i	Glob pattern(s) to include files
--include-tests, -T	Include TestXXX() functions (Go-specific)
--exclude, -e	Glob pattern(s) to exclude files
--exclude-internal, -E	Exclude 'internal' directories (Go-specific)	true
--match	Regular expression(s) to match identifiers
--symbol, -s	Symbol(s) to search for in code (TS/JS-specific)
--clear, -c	Force-clear comments in generation prompt (Go-specific)
--branch	Branch name to commit changes to (leave empty to not commit)
--limit	Limit the number of files to generate documentation for	0
--dry	Print the changes without applying them	false
--model, -m	OpenAI model used to generate documentation	"gpt-3.5-turbo"
--maxTokens	Maximum number of tokens to generate for a single documentation	512
--parallel, -p	Number of files to handle concurrently	4
--workers	Number of workers to use per file	2
--override, -o	Override existing documentation (Go-specific)
--key	OpenAI API key
--verbose, -v	Enable verbose logging	false

Screenshots

License

MIT