codepi

Persistent project intelligence for AI coding assistants — no duplicate code, smart context, token savings.

codepi remembers what's built in your project, detects duplicate code before you generate it, and injects smart context into every AI prompt — so your AI assistant never builds something twice.

The Problem

AI coding assistants have no memory of your project between conversations. Every time you ask "build a login form" or "add authentication", the AI starts from scratch — wasting tokens, time, and often rebuilding things that already exist.

The Solution

codepi sits between you and your AI assistant. It:

• Scans your project and stores file summaries + embeddings in a local SQLite database
• Remembers every task completed and architectural decision made
• Detects when you're about to build something that already exists (via semantic similarity)
• Injects context into every AI prompt — what's built, what's related, what patterns to follow

Prerequisites

• Node.js 18 or later
• API key (free) from one of these providers:
  • NVIDIA (recommended) — free tier at build.nvidia.com
  • Anthropic — console.anthropic.com
  • OpenAI — platform.openai.com
• Ollama (optional) — local AI, no API key needed: ollama.com

Installation

npm install -g @pravin-sk/codepi

Quick Start

1. Get a free API key (pick one)

| Provider | Free Tier | Link | |---|---|---| | NVIDIA | Yes — generous free tier | build.nvidia.com | | Anthropic | Limited free credits | console.anthropic.com | | OpenAI | Requires billing | platform.openai.com | | Ollama | Free — runs locally | ollama.com |

2. Add key to your project

Create a .env file in your project folder:

NVIDIA_API_KEY=your_key_here

3. Start using codepi

cd my-project
codepi init
codepi scan
codepi ask "Explain the architecture of this project"
codepi check "authentication middleware"
codepi status

No Ollama required. Works immediately with any free API key.

Commands

| Command | Description | |---|---| | codepi init | Initialize codepi memory in the current project | | codepi scan | Scan all project files, generate embeddings, store in memory | | codepi status | Show scanned files, completed tasks, duplicates, token savings | | codepi ask <prompt> | Ask AI with full project context injected | | codepi check <prompt> | Check if a feature already exists via semantic search | | codepi duplicates | Find duplicate code across your project | | codepi history | Show past prompts and AI responses |

Examples

# Initialize
codepi init

# Scan (re-scans and replaces old data)
codepi scan

# Check before building
codepi check "user authentication system"

# Ask a question
codepi ask "What patterns do we use for error handling?"

# View history
codepi history

# Find duplicates
codepi duplicates

Environment Variables

Create a .env file in your project root:

| Variable | Default | Description | |---|---|---| | NVIDIA_API_KEY | — | NVIDIA free API key (recommended) | | AI_PROVIDER | auto-detect | Force provider: nvidia, openai, anthropic, ollama | | AI_MODEL | per-provider default | Override model name | | AI_API_KEY | — | Fallback API key for nvidia/openai/anthropic | | OPENAI_API_KEY | — | OpenAI API key | | ANTHROPIC_API_KEY | — | Anthropic API key | | AI_BASE_URL | — | Custom API endpoint | | AI_TIMEOUT | 300000 | Request timeout in milliseconds (default: 5 min) | | OLLAMA_URL | http://localhost:11434 | Ollama server URL |

Supported AI Providers

| Provider | Default Model | Free Tier | |---|---|---| | Ollama (local) | llama3.2 | Yes — fully local | | NVIDIA | meta/llama-3.1-8b-instruct | Yes — free API key | | OpenAI | gpt-4o | No — requires billing | | Anthropic | claude-sonnet-4-20250514 | No — requires billing |

How It Works

┌─────────────┐     ┌──────────────┐     ┌─────────────┐
│  codepi     │────▶│  File       │────▶│  SQLite     │
│  scan       │     │  Scanner    │     │  Memory DB  │
└─────────────┘     └──────────────┘     └─────────────┘
                                               │
┌─────────────┐     ┌──────────────┐          │
│  codepi     │────▶│  Context    │◀─────────┘
│  ask        │     │  Builder    │
└─────────────┘     └──────┬───────┘
                           │
                    ┌──────▼───────┐
                    │  AI Provider │
                    │  (Ollama /   │
                    │   OpenAI /   │
                    │   Anthropic  │
                    │   / NVIDIA)  │
                    └──────────────┘

1. codepi scan recursively walks your project, reads file contents, detects language, generates vector embeddings, and stores everything in a local SQLite database (.codepi/memory.db).

2. codepi ask builds a context block from memory — listing built files, related files (via embedding similarity), completed tasks, and architectural decisions — then sends it alongside your prompt to the configured AI provider.

3. codepi check compares your prompt against all file embeddings using cosine similarity, returning files that match above a configurable threshold.

4. codepi duplicates compares every file embedding against every other file, flagging pairs with similarity above 85%.

5. Token tracking records prompt tokens and estimated savings — showing you how much context you're saving versus sending the full codebase.

Tech Stack

• TypeScript — Full type safety
• SQLite (via better-sqlite3) — Local persistent memory
• Transformers.js (via @xenova/transformers) — Local embeddings
• Commander.js — CLI framework
• Chalk — Terminal styling

Contributing

1. Fork the repository
2. Create a feature branch
3. Run npm run build to verify compilation
4. Submit a pull request

License

MIT © PRAVIN SK