NightCode

AI-powered code assistant — multi-provider, extensible via MCP, with real-time tool feedback.

---

Overview

NightCode is a full-stack, locally-run AI coding assistant. Connect your own API keys, choose from 14 supported providers, and let the agent work with your codebase through a rich tool system. It supports file operations, shell commands, code search, artifact rendering, image generation, and extensible MCP servers.

---

Features

Multi-Provider AI

Bring your own keys. NightCode supports 14 providers with automatic failover, health monitoring, rate-limit tracking, and key rotation:

Provider	Identifier	Endpoint
Google Gemini	google	generativelanguage.googleapis.com
OpenAI	openai	api.openai.com
Groq	groq	api.groq.com
OpenRouter	openrouter	openrouter.ai
OpenCode	opencode	opencode.ai
Ollama	ollama	ollama.com
DeepSeek	deepseek	api.deepseek.com
xAI	xai	api.x.ai
Xiaomi	xiaomi	api.xiaomimimo.com
Cerebras	cerebras	api.cerebras.ai
Naga	naga	api.naga.ac
SambaNova	sambanova	api.sambanova.ai
FreeTheAI	freetheai	api.freetheai.xyz
Cloudflare	cloudflare	api.cloudflare.com

22 Built-in Tools

Tool	Purpose
read_file / write_file / create_file	Read, write, and create files
edit_file	Targeted file edits
delete_file / create_folder	File system management
list_directory / search_files / grep	Codebase navigation and search
shell / execute_workspace_script	Command execution
delegate_task	Sub-agent delegation
ask	Interactive user questions
create_artifact / read_artifact / edit_artifact / list_artifacts	Rich content rendering
generate_image	AI image generation (Google Gemini)
search_memories	Agent memory retrieval
task	Task tracking
plan_exit	Plan completion signaling
get_errors / run_tests	Build and test integration

MCP (Model Context Protocol)

Extend NightCode with any MCP-compatible server. Plug in database tools, API clients, browser automation, or any custom capability:

{
  "name": "my-server",
  "type": "local",
  "command": "python my_server.py",
  "args": [],
  "enabled": true
}

MCP tools are automatically merged into the agent's available toolset with server-prefixed names.

Skills System

Inject specialized instructions into the agent's context. Enable skills globally from the customize page or activate them per-message with @skill-name in the prompt input.

Rich Artifacts

Render code, markdown, HTML, SVG, Mermaid diagrams, and math equations inline — streamed token-by-token for instant feedback.

Smart Context Management

Automatic context window compaction, token counting, and sliding-window decisions prevent context overflow while preserving critical information.

---

Architecture

Message → runEngine() → Adapter → Provider Stream (text + tool_calls)
                              ↓
                    dispatchTool() → execute → verify
                              ↓
                    Feed result back → repeat until done
                              ↓
                    ContextManager (auto-compaction)
                              ↓
                    Cache + Telemetry

NightCode uses a dual-engine architecture. The new event-driven engine (lib/engine2/) handles streaming, tool dispatch, context management, and caching. The legacy engine (lib/engine/) remains for backward compatibility.

---

Quick Start

Prerequisites

• Node.js 20+
• pnpm (or npm)
• At least one AI provider API key

Setup

# Clone the repository
git clone https://github.com/CodewithMubasher/NightCode3.git
cd NightCode3

# Install dependencies
pnpm install

# Configure your API keys
cp .env.example .env.local
# Edit .env.local with your keys

Run

pnpm dev        # Development server
pnpm build      # Production build
pnpm start      # Start production server

Open http://localhost:3000 in your browser.

Other Commands

Command	Description
pnpm lint	Run ESLint
pnpm format	Format code with Prettier
pnpm typecheck	TypeScript type checking

---

Configuration

Environment Variables

All provider keys go in .env.local:

# At least one required
OPENAI_API_KEY=sk-...
GOOGLE_GENERATIVE_AI_API_KEY=AIza...
GROQ_API_KEY=gsk_...

# Optional: image generation
GOOGLE_IMAGE_KEY_1=AIza...
GENERATE_IMAGE_MODEL=gemini-2.5-flash-image

# Optional: restrict file operations
BUILD_WORKSPACE=C:\Users\me\project

Keys are stored in SQLite at runtime — env vars serve as fallback defaults.

MCP Servers

Copy the example and configure your servers:

cp mcp-servers.example.json mcp-servers.json

Each server needs a name, type (local or remote), command, and optional environment variables.

Settings

Configure theme (dark/light/system), primary color, default model/provider, temperature, max tokens, and more from the Settings dialog in the app.

---

Project Structure

app/                  # Next.js App Router pages and API routes
├── (dashboard)/      # Main UI: chat, projects, customize
└── api/              # API endpoints (chat, keys, MCP, skills, etc.)

components/           # React UI components
├── ui/               # shadcn/ui primitives
├── chat/             # Chat interface
└── ai-elements/      # AI-rendered content (canvas, artifacts, etc.)

lib/                  # Core logic
├── engine/           # Legacy engine
├── engine2/          # New event-driven engine
├── mcp/              # MCP integration
├── db/               # SQLite database
└── keys/             # API key routing

store/                # Zustand state management
types/                # TypeScript type definitions
.skills/              # Skill definition files

---

Tech Stack

Layer	Technology
Framework	Next.js 16 (App Router)
UI	React 19, TailwindCSS v4, shadcn/ui
State	Zustand 5 with persist
AI SDK	Vercel AI SDK 6, LangChain
LLM Proviers	OpenAI, Google, Groq, OpenRouter, Ollama, +8 more
MCP	Model Context Protocol SDK
Database	better-sqlite3
Content	Streamdown, KaTeX, Mermaid, xFlow
Icons	Lucide
Animations	Motion (Framer Motion)

---

License

MIT — see LICENSE.

Copyright © 2026 Mubasher Chaudhary.