58e10a5a95
- Add POST /api/library/rename/:item_type/:name endpoint supporting skills, commands, rules, agents, tools, and workspace templates - Implement dry_run mode to preview changes before applying - Auto-update all cross-references in related configs and workspaces - Add RenameDialog component with preview and apply workflow - Integrate rename action into Skills, Commands, and Rules config pages - Fix settings page to sync config before restarting OpenCode - Clarify INSTALL.md dashboard deployment options (Vercel vs local) - Add docs-site scaffolding (Nextra-based documentation)
101 lines
5.8 KiB
Plaintext
101 lines
5.8 KiB
Plaintext
---
|
|
title: Open Agent
|
|
description: Self-hosted control plane for autonomous AI agents
|
|
---
|
|
|
|
import { Callout } from 'nextra/components'
|
|
|
|
<Callout type="info">
|
|
**Are you an AI?** Check [/llms.txt](/llms.txt) or add `.md` to any page URL to get raw markdown and save tokens.
|
|
</Callout>
|
|
|
|
# Open Agent
|
|
|
|
Run AI coding agents on your own server. Start a task, close your laptop, check back tomorrow.
|
|
|
|
## The Problem
|
|
|
|
You're using Claude Code or Cursor. It's good. But:
|
|
|
|
- **Sessions time out.** That 4-hour refactor? Dead at 80%.
|
|
- **Your code goes somewhere.** Fine for side projects. Not fine for work.
|
|
- **You can't customize it.** Same generic behavior for every project.
|
|
|
|
Open Agent fixes all three.
|
|
|
|
## What It Actually Does
|
|
|
|
Open Agent is a control plane that runs on your server. You connect to it from a web dashboard, iOS app, or CLI. You tell an agent what to do. It works until it's done, whether that takes 10 minutes or 10 hours.
|
|
|
|
Your code stays on your machine. You define how the agent behaves through a git repo of skills, commands, and rules. Each project gets its own isolated container.
|
|
|
|
Under the hood, it runs [OpenCode](https://opencode.ai) for the actual agent work. Open Agent handles everything around it: starting tasks, streaming progress, managing environments, syncing configuration.
|
|
|
|
## When To Use This
|
|
|
|
**You have a task that takes hours.**
|
|
Point an agent at a GitHub issue before bed. Wake up to a PR. Review the diff, not the process.
|
|
|
|
**Your code can't leave your machines.**
|
|
Healthcare, finance, defense, or just "my company said no." Run local inference or route API calls through your own infrastructure.
|
|
|
|
**You want agents that know your codebase.**
|
|
Write skills that encode your team's patterns. "When touching the payments module, always run the fraud detection tests." The agent follows them.
|
|
|
|
**You run multiple projects with different needs.**
|
|
Container workspaces isolate everything. Your Node project doesn't see your Python project. Different tools, different rules, no cross-contamination.
|
|
|
|
## How It Works
|
|
|
|
```
|
|
┌─────────────────────────────────────────────────────────┐
|
|
│ Your Devices │
|
|
│ ┌─────────────┐ ┌─────────────┐ ┌─────────────────┐ │
|
|
│ │ Web Dashboard│ │ iOS App │ │ Claude Code CLI │ │
|
|
│ └──────┬──────┘ └──────┬──────┘ └────────┬────────┘ │
|
|
└─────────┼────────────────┼──────────────────┼───────────┘
|
|
│ │ │
|
|
└────────────────┼──────────────────┘
|
|
│ HTTPS
|
|
▼
|
|
┌─────────────────────────────────────────────────────────┐
|
|
│ Your Server (~$30/mo) │
|
|
│ ┌─────────────────────────────────────────────────┐ │
|
|
│ │ Open Agent │ │
|
|
│ │ • Starts and monitors agent tasks │ │
|
|
│ │ • Manages isolated container workspaces │ │
|
|
│ │ • Syncs skills and rules from your git repo │ │
|
|
│ └──────────────────────┬──────────────────────────┘ │
|
|
│ │ │
|
|
│ ┌──────────────────────▼──────────────────────────┐ │
|
|
│ │ OpenCode │ │
|
|
│ │ • Runs the actual agent logic │ │
|
|
│ │ • Executes tools, writes code │ │
|
|
│ │ • Talks to Claude/GPT/local models │ │
|
|
│ └─────────────────────────────────────────────────┘ │
|
|
└─────────────────────────────────────────────────────────┘
|
|
```
|
|
|
|
Open Agent doesn't do AI. It manages the infrastructure around AI. OpenCode does the thinking. Open Agent makes sure it has somewhere to run, something to work on, and a way for you to watch.
|
|
|
|
**Note**: The setup guide installs the **backend** on your server. For the dashboard, you have options:
|
|
- **Vercel**: Host the Next.js dashboard for free
|
|
- **Local**: Run `bun dev` in the dashboard folder on your machine
|
|
- **iOS**: Download the app and enter your server URL
|
|
|
|
## Setup Takes 5 Minutes
|
|
|
|
If you have a server and a domain, tell Claude or Cursor:
|
|
|
|
> "Deploy Open Agent on my server at `1.2.3.4` with domain `agent.example.com`. Read INSTALL.md for the full guide."
|
|
|
|
It handles systemd, nginx, SSL, everything. Or [do it manually](/setup) if you prefer.
|
|
|
|
Don't have a server? Get one from [Hetzner](https://www.hetzner.com/) or DigitalOcean for ~$30/month.
|
|
|
|
## Next
|
|
|
|
1. **[Setup](/setup)** to get Open Agent running on your server
|
|
2. **[First Mission](/first-mission)** to run your first agent task
|
|
3. **[Library](/library)** to customize how your agents behave
|