v1.0 — Production Release

The AI Agent
That Builds Its
Own Tools

Privacy-first, self-evolving AI agent with GraphRAG memory, auto skill generation, plugin ecosystem, and web dashboard. 100% local. Zero config.

Privacy-First Zero Config Self-Evolving MIT Licensed

0

Core Modules

0

+ Test Cases

0

CLI Commands

0

% Open Source

Features

Packed with Power

Everything you need for a powerful, local AI agent that learns and evolves with every interaction.

🧠

GraphRAG Memory

Persistent knowledge graph using NetworkX for intelligent context retrieval across sessions.

🔧

Auto Skill Generation

The agent writes its own tools during runtime and saves them permanently for reuse.

🔌

Plugin System

Extend with custom plugins, hot-reload support, and a hook-based architecture.

🌐

Web Dashboard

Optional FastAPI dashboard for monitoring skills, memory graph stats, and task history.

🔒

Privacy-First

Local-first execution with Ollama. Your data stays on your machine. Optional cloud models via LiteLLM.

📦

Sandboxed Execution

Isolated skill execution with subprocess isolation, timeouts, and memory limits.

🤖

Multi-Agent System

Orchestrate multiple agents with task delegation, role-based routing, and shared memory.

🛡️

Enterprise Security

Encrypted cloud sync (Fernet), audit logging with RBAC, and structured JSON-lines logs.

🎯

Multi-Model

Works with any model via LiteLLM — Ollama, OpenAI, Anthropic, Google, and more.

How It Works

Four Steps to Intelligence

From zero to a self-evolving AI agent in minutes.

1

Install & Run

Install via pip, pull an Ollama model, and run your first task. No configuration files, no API keys, no setup wizards.

2

Agent Processes & Remembers

The agent processes your prompt using GraphRAG memory for context-aware responses. Every interaction enriches the knowledge graph.

3

Skills Auto-Generate

When the agent needs a tool it doesn't have, it writes one and saves it permanently. Future requests reuse saved skills automatically.

4

Evolves Over Time

With each interaction, the knowledge graph grows and the skill tree expands. Your agent gets smarter the more you use it.

Quick Start

Up and Running in 60 Seconds

terminal
$ pip install nexus-agent
$ ollama pull llama3
$ nexus run "Create a Python web scraper"

⚡ NexusAgent initialized. Model: ollama/llama3
🧠 Consulting GraphRAG memory...

Agent Response:
[NEW SKILL: web_scraper]
```python
import requests
from bs4 import BeautifulSoup
def scrape(url): ...
```

$ nexus skills
📋 Skill Tree: web_scraper ✅

$ nexus evolve
⚡ Evolving... Memory graph: 47 nodes, 123 edges
Architecture

How It's Built

A modular architecture designed for extensibility, privacy, and performance.

Core Engine

NexusAgent Core orchestrator & LiteLLM router
GraphMemory NetworkX-based GraphRAG store
SkillTree Auto-generated tool manager
Sandbox Isolated subprocess execution
🔌

Extensions

Plugins Hot-reload hook system
Web Dashboard FastAPI monitoring UI
Multi-Agent Task routing & collaboration
Cloud Sync Fernet-encrypted sync engine
🔄

Data Flow

CLI / Web NexusAgent Core GraphRAG + Skills LiteLLM Response
Comparison

Why NexusAgent?

Based on publicly available documentation as of April 2026. Verify for your use case.

Feature NexusAgent Aider Continue.dev Cursor OpenHands
Runs Locally (with Ollama)
GraphRAG Memory
Self-Evolving Skills
Plugin System
Zero-Config Setup⚠️⚠️⚠️
CLI Interface
Sandboxed Execution
Open SourceMITApacheApacheProprietaryMIT
Principles

Built on Principles

🔒

Privacy by Design

No telemetry, no tracking, no data leaves your machine when using local models. Your code stays yours.

🌱

Self-Improving

The more you use it, the more capable it becomes. Skills accumulate and the knowledge graph grows organically.

🔓

Open & Extensible

MIT licensed with a plugin system. Modify, extend, and share freely. No vendor lock-in.

Roadmap

The Journey So Far

v0.1 — Core Agent

GraphRAG memory (NetworkX), auto skill generation, CLI, LiteLLM + Ollama

v0.2 — Plugins & Dashboard

Config management, plugin system with hot-reload, FastAPI dashboard, sandboxed execution, export, Docker, CI/CD

v0.3 — Multi-Agent

Orchestration engine, task delegation & routing, collaborative shared memory, agent roles, priority queue

v0.4 — Voice & IDE

Voice interface (Whisper STT + TTS), IDE integration (JSON-RPC), AST-aware code memory, context window management

v1.0 — Production

Encrypted cloud sync, audit logging & RBAC, skill marketplace, performance benchmarks, mobile companion API, 140+ tests

FAQ

Common Questions

Is my data sent anywhere?
By default, NexusAgent uses local Ollama models — no data leaves your machine. If you configure cloud providers (OpenAI, Anthropic), only prompts are sent to those APIs. No additional telemetry or tracking is collected.
What models does it support?
Any model supported by LiteLLM: Ollama (Llama, Mistral, CodeLlama, etc.), OpenAI (GPT-4), Anthropic (Claude), Google (Gemini), and dozens more. Configure with nexus config set model.default <model>.
How does self-evolution work?
When the agent encounters a task requiring a tool it doesn't have, it generates the tool code and saves it permanently to the skill tree. Future similar requests reuse the saved skill instead of regenerating code. The nexus evolve command scans your workspace to enrich the knowledge graph.
Can I use it without Ollama?
Yes. Configure any LiteLLM-compatible provider via nexus config set model.default openai/gpt-4. You'll need an API key set as an environment variable for cloud providers.
How do I write plugins?
Create a Python file in .nexus/plugins/ with functions prefixed nexus_*. These hooks are called automatically at various points in the agent lifecycle. See the plugin guide for details.
Does it work on Windows?
Yes. NexusAgent is cross-platform and works on Windows, macOS, and Linux. Sandboxed execution uses sys.executable for compatibility across platforms.
Documentation

Learn More

Everything you need to get the most out of NexusAgent.

Ready to Build?

Get started with NexusAgent in under 2 minutes.

$ pip install nexus-agent