# Agenvoy

> A self-hosted AI agent harness that runs on your own machine. A single Go daemon talks to 10 LLM providers through one interface, writes and sandbox-tests its own tools when nothing existing covers a request, and exposes its whole tool library to other agents over MCP. Reaches you through TUI, browser, Telegram and Discord. Built in Taiwan by Pardn Chiu, Apache 2.0.

Docs are also available in Traditional Chinese under https://agenvoy.com/zh/docs/ (same slugs).

## Overview

- [Home](https://agenvoy.com/docs/): Agenvoy documentation — a personal AI agent that runs on your machine. Guides for setup, tools, MCP, memory, scheduling, and security.
- [Quickstart](https://agenvoy.com/docs/quickstart): Beginner walkthrough — one-line install, a free NVIDIA NIM key, and gpt-oss-20b running in about five minutes.
- [Getting Started](https://agenvoy.com/docs/getting-started): Install Agenvoy and run your first AI agent session in under 60 seconds. One command setup for macOS and Linux.

## Concepts

- [Sessions & Agents](https://agenvoy.com/docs/sessions): Manage sessions, agent personas, routing rules, and per-session concurrency in Agenvoy.
- [Execution Engine](https://agenvoy.com/docs/execution-engine): How the Agenvoy iteration loop, three-pass tool dispatch, and circuit breaker work under the hood.
- [Providers](https://agenvoy.com/docs/providers): Configure 10 LLM providers — OpenAI, Claude, Gemini, Grok, Copilot, DeepSeek, Mistral, NVIDIA NIM, OpenRouter and Cloudflare — plus a local/custom endpoint.

## User Guide

- [CLI Commands](https://agenvoy.com/docs/cli-commands): Agenvoy CLI commands, make shortcuts, input prefixes, and environment variables reference.
- [TUI Guide](https://agenvoy.com/docs/tui-guide): Agenvoy TUI keyboard shortcuts, slash commands, and interactive session management.
- [REST API](https://agenvoy.com/docs/rest-api): OpenAI-compatible REST API endpoints for chat completions, sessions, and log replay.
- [Configuration](https://agenvoy.com/docs/config-files): Agenvoy configuration file layout — config.json, bot.md, permission modes, and runtime limits.
- [Integration Config](https://agenvoy.com/docs/config-integrations): Configure MCP servers, LLM providers, KuraDB, Telegram bot, and Discord bot integrations.

## Tools

- [Built-in Tools](https://agenvoy.com/docs/built-in-tools): Built-in AI agent tools — file operations, web search, orchestration, memory, RAG, scheduling, and media.
- [Tool Extension](https://agenvoy.com/docs/tool-extension): Auto-generate custom tools from natural language. Add script, API, or MCP tools to extend Agenvoy.
- [Tool Design & Rules](https://agenvoy.com/docs/tool-rules): Tool design guidelines — concurrency markers, timeouts, credential auto-heal, and naming conventions.

## Features

- [Memory System](https://agenvoy.com/docs/memory-system): Three-tier conversation memory — rolling context window, semantic vector search, and FTS5 SQLite archive.
- [Skill System](https://agenvoy.com/docs/skill-basics): Create loadable markdown skill packs with YAML frontmatter, slash-command and natural-language triggers.
- [Scheduler](https://agenvoy.com/docs/scheduler-skills): Cron and one-shot task scheduling with skill binding, fsnotify hot-reload, and revision-based recovery.
- [MCP Server](https://agenvoy.com/docs/mcp-server): Run Agenvoy as an MCP server — expose sandboxed tools to Claude Code, Codex, Cursor, and OpenCode.
- [MCP Client](https://agenvoy.com/docs/mcp-client): Connect Agenvoy to external MCP servers via stdio or HTTP/SSE with auto-discovery and hot-reload.
- [KuraDB & RAG](https://agenvoy.com/docs/kuradb-rag): Enable KuraDB for keyword and semantic document search (RAG) over your personal knowledge base.

## Security

- [Sandbox](https://agenvoy.com/docs/sandbox): OS-native command sandbox — bubblewrap on Linux, sandbox-exec on macOS. CPU, memory, and network limits.
- [Security Model](https://agenvoy.com/docs/security): Agenvoy security model — permission modes, macOS Keychain, system prompt protection, and MCP isolation.

## Reference

- [Architecture](https://agenvoy.com/docs/architecture): Agenvoy architecture — system layers, daemon lifecycle, cross-cutting principles, and TUI design.
- [Comparison](https://agenvoy.com/docs/comparison): Compare Agenvoy vs Claude Code, Codex CLI, Cursor, Aider, and other AI agent platforms.

## Optional

- [Release Notes](https://agenvoy.com/docs/released/): All Agenvoy release notes — changelogs, features, and fixes by version.
- [Source Repository](https://github.com/agenvoy/Agenvoy): Go source, issues, and Apache 2.0 license.
- [Go Reference](https://pkg.go.dev/github.com/pardnchiu/agenvoy): Generated package documentation on pkg.go.dev.