Agentbot

Docker for AI workers.

Build, compose, and deploy autonomous AI agents. Open-source infrastructure for the agent economy.

curl -fsSL agentbot.sh/install | bash

One command. Your agent is running on your machine — connected to Telegram, Discord, or WhatsApp. Your API key talks directly to your LLM provider. We don't touch the costs.

npx agentbot init     # Create your first agent
npx agentbot dev      # Run locally
npx agentbot deploy   # Ship to production

→ Agentbot Cloud for managed deployment

What is an Agentbot Agent?

An agent is a Markdown file with YAML frontmatter:

---
name: researcher
description: Deep research agent for web analysis
model: openrouter/anthropic/claude-3.5-sonnet
tools: [bash, read, write, web, think, memory]
permissions:
  bash: dangerous
  read: safe
  write: dangerous
---

# Researcher Agent

You are a deep research agent specializing in web analysis.

That's it. Define your agent, give it tools, set permissions, write its instructions.

Quick Start

# Create your first agent
npx agentbot init

# Run locally
npx agentbot dev

# Deploy to Agentbot Cloud
npx agentbot deploy

Packages

Package	Description
@agentbot/sdk	Core SDK — agent parser, orchestration, plugins, types
@agentbot/cli	Developer CLI — init, dev, run
@agentbot/plugins	Official integrations — Discord, Telegram, email, browser

Plugins

Extend agents with capabilities:

import { discordPlugin, telegramPlugin } from '@agentbot/plugins'

// Plugins provide tools agents can use
const agent = new Agent({
  tools: ['discord_send', 'telegram_send', 'email_send', 'browser_navigate']
})

Plugin	Tools
Discord	discord_send, discord_read
Telegram	telegram_send, telegram_webhook
Email	email_send, email_template
Browser	browser_navigate, browser_scrape, browser_screenshot

Build your own plugin:

import { definePlugin, defineTool } from '@agentbot/sdk'

export const myPlugin = definePlugin({
  name: 'my-integration',
  description: 'Custom integration',
  version: '0.1.0',
  tools: [
    defineTool({
      name: 'my_tool',
      description: 'Does something useful',
      parameters: { input: { type: 'string', required: true } },
      async execute(args, ctx) {
        ctx.log(`Running with: ${args.input}`)
        return { result: 'done' }
      }
    })
  ]
})

Examples

• Research Agent — Web research and analysis
• Outreach Agent — Lead generation and cold messaging
• Content Agent — Blog posts, social media, documentation
• Crypto Analyst — Autonomous 24/7 market scanner
• Barista Agent — Morning motivation (5 AM survival mode)
• Multi-Agent Workflow — Compose agents into pipelines

Architecture

Agent Definition (.md)
    ↓
Agent Runtime (SDK)
    ↓
Tool Orchestration (parallel reads, serial writes)
    ↓
Memory + Hooks
    ↓
Deploy (self-host or Agentbot Cloud)

Agent Spec

Frontmatter Fields

Field	Type	Required	Description
name	string	✅	Unique agent identifier
description	string	✅	Human-readable description
model	string	✅	LLM model to use
tools	string[]	✅	Available tools
permissions	object	❌	Per-tool permission overrides

Permission Levels

• safe — Auto-approve (reads, searches, memory lookups)
• dangerous — Require user approval (writes, shell commands)
• destructive — Block entirely (rm -rf, DROP TABLE)

Built-in Tools

• bash — Shell command execution
• read / write — File system operations
• web — Web search and fetching
• think — Internal reasoning
• memory — Persistent memory storage

Orchestration

Agentbot automatically parallelizes read-only tool calls and serializes writes:

[read] [grep] [web_search] → parallel (Promise.all)
    ↓
[write file] → serial (wait for completion)
    ↓
[bash test] → serial (wait for completion)
    ↓
[read] [memory_search] → parallel

This gives you maximum throughput without race conditions.

Verify Your Agent

Link your agent to a verified human identity via SelfClaw:

# Via dashboard (easiest)
open https://agentbot.sh/dashboard/verify

# Or directly
open https://selfclaw.ai/verify

Verified agents get a ✅ badge and are eligible for the marketplace.

→ Verification Guide

Self-Host

# Clone and run locally
git clone https://github.com/Eskyee/agentbot-sdk
cd agentbot-sdk
npm install
npm run build

# Run with Docker
OPENROUTER_API_KEY=sk-... docker compose up

# Or run directly
npx agentbot run examples/research-agent/agent.md

Agentbot Cloud

For production deployments with managed infrastructure:

• One-click deploy
• Hosted memory
• Dashboards and logs
• Team workspaces
• Secret storage
• Production uptime
• Premium model routing

→ agentbot.sh

Contributing

We welcome contributions! See CONTRIBUTING.md for guidelines.

Ways to Contribute

• Build agents and share them
• Create plugins for new integrations
• Improve the SDK and CLI
• Write documentation
• Report bugs

Ecosystem

Agentbot is part of a growing ecosystem:

• $AGENTBOT — Community token on Base and Solana
• Agentbot Cloud — Managed deployment platform
• baseFM — Onchain radio platform

License

MIT

---

Agentbot — Docker for AI workers. Build once, deploy anywhere.