Skip to content

Eskyee/agentbot-sdk

Repository files navigation

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

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:

License

MIT


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

About

Docker for AI workers. Build, compose, and deploy autonomous AI agents.

Resources

License

Contributing

Stars

Watchers

Forks

Releases

No releases published

Packages

 
 
 

Contributors