Developer Infrastructure · MIT License

Operating system
for AI agents.

Six production-ready primitives — filesystem, database, memory, network, processes, and events — secured and isolated per agent. Ship your agent in 5 minutes, not 5 weeks.

Start building free →View on GitHub ↗

6 primitives

5 min setup

MIT License

90d token TTL

The Problem

Stop reinventing infrastructure

Every agent team rebuilds the same boilerplate. AgentOS ships it once.

WITHOUT AGENTOS

before.ts

// Before AgentOS
const redis = new Redis(process.env.REDIS_URL);
const supabase = createClient(url, key);
const { exec } = require('child_process');

// Set up auth, rate limiting, sandboxing...
// Handle errors, timeouts, quotas...
// Write 500+ lines of infra code
// before a single line of agent logic.

WITH AGENTOS

after.ts

// With AgentOS
const os = new AgentOS({ apiKey: process.env.AGENT_OS_KEY });

await os.mem.set('key', value);
await os.db.insert('table', row);
await os.proc.execute(code, 'python');
// Auth, isolation, quotas — included.

Primitives

6 primitives, infinite possibilities

Everything an agent needs to read, write, compute, and communicate.

os.mem

Memory

Redis-backed key-value store with TTL, namespaced per agent.

mem_setmem_getmem_deletemem_listmem_incr

os.fs

Filesystem

Cloud file storage backed by Supabase. Each agent gets isolated storage.

fs_readfs_writefs_listfs_deletefs_mkdir

os.db

Database

PostgreSQL with per-agent schema isolation. Queries, transactions, DDL.

db_querydb_insertdb_updatedb_deletedb_transaction

os.net

Network

Outbound HTTP with SSRF protection, domain allowlisting, rate limiting.

net_http_getnet_http_postnet_http_putnet_dns_resolve

os.proc

Process

Sandboxed code execution: Python, JavaScript, Bash. Timeouts & quotas.

proc_executeproc_scheduleproc_spawnproc_kill

os.events

Events

Redis pub/sub messaging. Publish, subscribe, coordinate across agents.

events_publishevents_subscribeevents_list_topics

Live Example

See all 6 in 40 lines

A trading agent that monitors BTC and broadcasts buy signals.

trading-agent.ts

TypeScript

import { AgentOS } from '@agentos/sdk';

const os = new AgentOS({
  apiUrl: 'https://agentos-app.vercel.app',
  apiKey: process.env.AGENT_OS_KEY
});

// Monitor Bitcoin price
const price = await os.net.http_get(
  'https://api.binance.com/api/v3/ticker/price?symbol=BTCUSDT'
);

// Cache for 60 seconds
await os.mem.set('btc_price', price.data.price, 60);

// Store in database
await os.db.insert('prices', {
  symbol: 'BTC',
  price: parseFloat(price.data.price),
  timestamp: Date.now()
});

// Run analysis
const signal = await os.proc.execute(`
import numpy as np
prices = ${JSON.stringify(priceHistory)}
rsi = calculate_rsi(prices)
print('BUY' if rsi < 30 else 'HOLD')
`, 'python');

// Publish event
if (signal.output === 'BUY') {
  await os.events.publish('trading.signals', {
    symbol: 'BTC',
    action: 'BUY',
    price: price.data.price
  });
}

// Save report
await os.fs.write('report.json', JSON.stringify(signal));

Use Cases