PaaS Free Tier¶

Alternative VPS — deploy via git push, ga perlu manage server.

PaaS (Platform-as-a-Service) cocok kalo lo ga mau urus systemd, nginx, dll. Kasih repo, mereka handle deployment.

Pros vs Cons PaaS¶

Pros: - Deploy via git push - Auto-scaling - Built-in monitoring & logs - HTTPS otomatis - No server management

Cons: - Resource lebih kebatas vs VPS - Vendor lock-in (perlu deploy ulang kalo pindah) - Cold starts (kalo idle) - Background workers biasanya bayar

Provider yang masih punya free tier¶

Railway¶

$5 credit/bulan (cukup buat ~hobby bot trigger-based)
1 GB RAM, 1 GB disk
Deploy via GitHub repo

Buat AI agent dengan tool execution: - Bisa, tapi ga continuous (credit habis dalam ~15 jam runtime/bulan) - Cocok buat bot scheduled (cron-based)

Render¶

1 web service free
512 MB RAM
Spin down setelah 15 menit idle = ga cocok buat bot

Buat AI agent: ❌ jangan, karena spin down.

Fly.io¶

3 VM gratis (256 MB each)
Persistent volume 3 GB
160 GB bandwidth

Buat AI agent: - Bisa kalo agent ringan - Pakai Dockerfile, deploy via flyctl deploy

Koyeb¶

1 nano instance gratis
256 MB RAM
Tidak spin down

Buat AI agent: - Cocok buat bot kecil yang continuous - Pakai Docker / Buildpack

Replit¶

Free tier dengan "Always-On" cuma bayar
Tanpa Always-On: spin down

Buat AI agent: - Free tier ga cocok (spin down) - Replit Hosting paid $7/bulan worth it kalo lo udah comfortable di Replit IDE

Vercel¶

Hobby tier free (untuk frontend)
Serverless functions (1M invocations/month free)
Bot continuous: ❌ serverless, max execution 10 detik per request

Buat AI agent: - Cuma cocok buat webhook handler yang fast - Ga cocok buat agentic loop yang multi-step

Cloudflare Workers¶

100K req/day free
10ms CPU per req
KV storage 100K read/day

Buat AI agent: - Cuma cocok buat webhook router super ringan - Ga cocok buat heavy logic

Deploy pattern PaaS¶

Pattern 1: Dockerfile¶

Dockerfile:

FROM python:3.12-slim
WORKDIR /app
COPY requirements.txt .
RUN pip install -r requirements.txt
COPY . .
CMD ["python", "main.py"]

requirements.txt:

python-telegram-bot==21.0
openai
python-dotenv

Deploy ke Fly.io:

flyctl auth login
flyctl launch  # interactive setup
flyctl deploy

Pattern 2: Procfile (Railway/Koyeb)¶

Procfile:

worker: python main.py

Auto-detected oleh platform.

Pattern 3: GitHub-based deploy¶

Push repo ke GitHub
Sign up Render/Railway
Connect GitHub
Pilih repo
Auto-deploy on push

Environment variable di PaaS¶

PaaS kasih dashboard buat set env var:

TELEGRAM_BOT_TOKEN=...
OPENAI_API_KEY=...

JANGAN commit .env ke repo. Tambah di .gitignore:

.env
credentials/
data/

Storage di PaaS¶

PaaS biasanya ephemeral filesystem — file yang lo bikin saat runtime ga persist setelah restart.

Solusi:

Option A: Persistent volume (kalo tersedia)¶

Fly.io: flyctl volumes create data --size 1

# fly.toml
[mounts]
  source = "data"
  destination = "/data"

Option B: Eksternal database¶

Pake managed DB:

Supabase (Postgres, 500MB free)
Neon (Postgres, 0.5GB free)
PlanetScale (MySQL, 5GB free)
MongoDB Atlas (512MB free)
Firebase Firestore (1GB free)

Adapt agent storage:

# Memory di Postgres
import psycopg2
conn = psycopg2.connect(os.environ["DATABASE_URL"])
cur = conn.cursor()
cur.execute("CREATE TABLE IF NOT EXISTS memory (id SERIAL, note TEXT, ts TIMESTAMP DEFAULT NOW())")
cur.execute("INSERT INTO memory (note) VALUES (%s)", (note_text,))
conn.commit()

Option C: S3 / R2 untuk file¶

import boto3
s3 = boto3.client('s3', endpoint_url=os.environ['S3_ENDPOINT'])
s3.put_object(Bucket='agent-data', Key='memory.json', Body=json.dumps(memory))

Cloudflare R2 free 10GB storage + 10M read/month = sangat cukup.

Latency considerations¶

Source	Target	Latency
Indonesia	Singapore (Oracle)	30-60ms
Indonesia	AWS us-east-1	250ms
Indonesia	GCP us-west1	200ms
Indonesia	Fly.io sin (Singapore)	30-60ms
Indonesia	Render Frankfurt	350ms

Telegram bot latency tolerant (user OK kalo respon 2-5 detik), tapi kalo lo bikin web UI, pilih region terdekat.

Cost projection¶

Skenario: AI agent jalan 24/7 setelah free tier habis.

Provider	Setup	Cost/bulan
Oracle Cloud	A1.Flex 4 OCPU 24GB	$0 forever (kalo ga di-suspend)
AWS	t3.micro after free tier	~$10
GCP	e2-micro after... still free	$0 forever (kalo policy ga ganti)
Hetzner	CX11 (paid)	$4.30
Fly.io	Free + paid volume	$2-5
Railway	Hobby plan	$5
Render	Standard (always-on)	$7
Koyeb	Standard	$5
DigitalOcean	$4/mo droplet	$4-6

Trade-off PaaS vs VPS untuk agent¶

Aspek	VPS (Oracle/AWS)	PaaS (Fly.io/Railway)
Setup time	2-3 jam	15-30 menit
Resource	Lebih besar	Lebih kecil
Cost long-term	Bisa $0 (Oracle)	Hampir selalu bayar
Maintenance	Manual (apt update, dll)	Otomatis
Storage	Persistent	Ephemeral (kecuali volume)
systemd / nginx	Setup manual	Built-in / abstracted
SSH access	Ya	Biasanya tidak (cuma logs)
Multi-platform	Bisa (banyak bot di 1 VPS)	Sulit (1 deploy per app)

Saran:

Start dengan VPS gratis (Oracle/AWS) kalo lo comfortable dengan terminal
PaaS cocok kalo lo ga mau ribet, atau bot lo simple

Hybrid setup¶

Yang sering gua rekomen:

Bot agent: VPS (Oracle/AWS) — long-running, full control
Web dashboard agent: Vercel — frontend statis + serverless API
Database: Supabase / Neon — managed Postgres, free tier
Storage: Cloudflare R2 — backup, file attachment

Cost total: $0/bulan untuk small scale.

Migration path¶

Kalo lo udah pake free tier dan butuh upgrade:

Oracle (free) → Oracle paid VM → AWS / GCP production

Atau:

PaaS (free) → PaaS paid → VPS (kalo butuh control)

Plan migration dari awal — pakai env var, jangan hardcode IP, simpan state external. Migration jadi gampang.