Developers

Getting Started

Introduction to Nebo Development

Nebo is an open-source personal AI assistant licensed under Apache 2.0. It runs locally on your computer — your data stays with you, your extensions are yours to build and share.

What You Can Build

Nebo's extension system is designed around a simple idea: the best way to make AI useful is to let people teach it new things. There are three ways to extend Nebo:

Skills

A Skill is a markdown file (SKILL.md) that teaches Nebo how to do something new. No code required. Write instructions in plain language, define triggers, and Nebo follows them.

Agents

An Agent is a personality and capability bundle. It includes an AGENT.md that defines how Nebo should behave, plus dependencies on skills and other agents. Think of it as a pre-configured agent persona.

Plugins

A Plugin is a shared native binary that multiple skills depend on. Build a CLI tool once, publish it to NeboLoop, and any skill can use it via environment variables. Plugins support authentication (OAuth flows), event production (NDJSON watch processes), and cross-platform distribution.

MCP Servers

Nebo supports the Model Context Protocol (MCP), letting you connect external tools and data sources. MCP servers give Nebo access to databases, APIs, file systems, and custom tooling — all through a standard protocol with AES-256-GCM encryption.

Design Philosophy

Nebo is built on a few core principles:

  • No code required. Most extensions are just markdown. If you can write instructions for a person, you can write a skill for Nebo.
  • Composable. Skills can be combined into agents. Agents can depend on other agents. Small pieces, loosely joined.
  • Local first. Extensions run on the user's machine. No cloud dependency, no latency penalty, no data leaving the device.
  • Open source. Nebo is Apache 2.0. Fork it, modify it, build on it. The marketplace is optional — you can side-load any extension.

Architecture Overview

Nebo is built in Rust with a workspace of focused crates:

Crate Purpose
ai Provider implementations (Anthropic, OpenAI, Gemini, Ollama)
tools Tool registry, policy engine, skills loader
agent Runner, session, memory, compaction
mcp MCP bridge with AES-256-GCM encryption
server Axum HTTP server, WebSocket handlers
db SQLite store with migrations

Next Steps

Architecture