NITIN AI 🤖

An autonomous AI software developer for your terminal. NITIN AI thinks, plans, writes code, runs tests, and learns your project — so you don't have to repeat yourself.

  ███╗   ██╗██╗████████╗██╗███╗   ██╗
  ████╗  ██║██║╚══██╔══╝██║████╗  ██║
  ██╔██╗ ██║██║   ██║   ██║██╔██╗ ██║
  ██║╚██╗██║██║   ██║   ██║██║╚██╗██║
  ██║ ╚████║██║   ██║   ██║██║ ╚████║
  ╚═╝  ╚═══╝╚═╝   ╚═╝   ╚═╝╚═╝  ╚═══╝

What makes it different

Most AI coding tools read your entire codebase on every prompt. NITIN AI builds a knowledge graph of your project — functions, classes, imports, dependencies — and only loads the files actually relevant to your task. This makes it fast, cheap, and accurate on large codebases.

Requirements

• Node.js 20+
• A Gemini API key (free at aistudio.google.com)
• A Groq API key (free at console.groq.com) (fallback)

Installation

git clone https://github.com/you/nitin-ai
cd nitin-ai
npm install
cp .env.example .env        # fill in your API keys
npm link                    # makes `nitin` available globally

Quick start

cd your-project
nitin init                  # scan project, build knowledge graph (~1-2 min)
nitin do "add a /health endpoint"
nitin do "fix the login bug"
nitin do "write tests for the auth module"
nitin                       # interactive REPL

Commands

| Command | Description | |---|---| | nitin | Interactive REPL mode | | nitin do "<task>" | One-shot autonomous task | | nitin init | Scan project and build knowledge graph | | nitin update | Refresh graph after manual changes | | nitin status | Show API quota and graph stats | | nitin undo | Rollback last change (git stash + graph restore) | | nitin config list | Show all settings | | nitin config set <key> <value> | Change a setting |

REPL commands

Inside nitin interactive mode:

| Command | Description | |---|---| | /model gemini | Force Gemini 2.0 Flash | | /model groq | Force Groq Llama 3.3 70B | | /model auto | Auto-select (default) | | /status | Show quota | | /help | Show commands | | /exit | Quit |

Configuration

nitin config set model auto          # auto | gemini | groq
nitin config set auto_test true      # run tests after every change
nitin config set auto_review false   # review before applying changes
nitin config set temperature 0.7     # LLM creativity (0-1)
nitin config set log_level debug     # debug | info | warn | error

How it works

User task
    │
    ▼
[Classify intent]
    │
    ├── feature  → feature-workflow  (design → write → test → heal)
    ├── bugfix   → bugfix-workflow   (diagnose → fix → verify)
    ├── refactor → refactor-workflow (surgical edits → verify)
    ├── test     → test-workflow     (generate + heal tests)
    ├── review   → review-workflow   (structured code review)
    └── other    → generic ReAct loop
                        │
                   [Query graph]     ← only loads relevant files
                        │
                   [Plan steps]      ← LLM produces JSON plan
                        │
                   [Execute + Reflect] ← tools + self-healing
                        │
                   [Verify tests]
                        │
                   [Update graph]

Project structure

nitin-ai/
├── bin/nitin.js              # CLI entry point
├── src/
│   ├── core/                 # ReAct agent loop
│   ├── graph/                # Knowledge graph (graphology)
│   ├── llm/                  # Gemini + Groq clients, smart router
│   ├── tools/                # file, shell, test tools
│   ├── workflows/            # feature / bugfix / refactor / test / review
│   ├── embeddings/           # chunker, embedder, vector store (pure-JS)
│   ├── ui/                   # renderer, diff viewer, progress
│   ├── commands/             # config, undo
│   ├── safety/               # command filter (BLOCKED / CONFIRM / SAFE)
│   └── utils/                # logger, tokenizer, config-loader, errors
└── .nitin/                   # Created in your project
    ├── graph.json            # Knowledge graph
    ├── embeddings.json       # Vector embeddings
    ├── quota.json            # API usage tracking
    ├── config.json           # Project settings
    └── checkpoints/          # Graph snapshots for undo

LLM routing

• Primary: Gemini 2.0 Flash — 1500 req/day, 1M tokens/day (free)
• Fallback: Groq Llama 3.3 70B — 30 req/min (free)
• Auto-switches on rate limit or quota exhaustion
• Quota persisted to .nitin/quota.json, resets at UTC midnight
• Manual override: nitin config set model groq

Safety

Every shell command is classified before execution:

• BLOCKED: fork bombs, rm -rf /, raw device writes — never run
• CONFIRM: sudo, rm, git push --force, SQL DROP — always prompts
• SAFE: reads, searches, npm test — auto-approved

File edits show a coloured diff and ask for approval before writing.

Running tests

npm test               # run all tests
npx vitest run         # same
npx vitest             # watch mode

License

MIT