The Toolkit

What you need to build with AI agents

This chapter lists every tool I used, how much it costs, and why I chose it. If you want to follow along with the 31-day sprint, this is your shopping list.

The Agent Platform: OpenClaw

OpenClaw is open-source software that turns an AI model into a persistent agent. It runs on your own server and connects to messaging apps.

What OpenClaw provides:

Messaging bridge. Connects the AI to Telegram, WhatsApp, Signal, Discord, and others. You message the agent like you would message a person.
Tool access. The agent can run shell commands, read and write files, browse the web, and call APIs.
Memory. The agent has persistent files it can read and write. These survive across conversations.
Skills. Modular instruction sets that teach the agent how to do specific things (check weather, manage GitHub issues, control smart home devices).
Sub-agents. The agent can spawn separate sessions to handle tasks in parallel.
Cron jobs. The agent can schedule tasks to run at specific times.

OpenClaw is free. You need a server to run it on and an API key from an AI provider (Anthropic, OpenAI, etc.).

Installation: Follow the instructions at docs.openclaw.ai. The basic setup takes about 30 minutes.

The AI Model: Claude by Anthropic

I use Claude as the language model behind Jarvis. Specifically:

Claude Opus for complex tasks (debugging, architecture decisions, writing)
Claude Sonnet for routine tasks (scaffolding, simple fixes, sub-agent work)

I initially bought $25 of Anthropic API credits and started using them directly. Within 15 minutes of real agent work, I had burned through $6–7. The API also has very low rate limits for new accounts — the agent would get throttled after a few back-to-back tasks.

So I switched to my Claude Code Max subscription ($200/month). This gives higher rate limits and more predictable costs. It still hits rate limits sometimes after 4–5 consecutive heavy tasks, but it recovers after a short cooldown period. For sustained agent work, a subscription plan is more practical than pay-per-token API credits.

You could use a different model. OpenClaw supports OpenAI (GPT-4), Google (Gemini), DeepSeek, and local models. I chose Claude because it is the best at following complex instructions and working with code, based on my testing.

The Server

A DigitalOcean droplet:

Size: 2 vCPUs, 4GB RAM
OS: Ubuntu 22.04
Cost: $24/month
Location: Bangalore (BLR1) — close to me for low latency

This is enough to run OpenClaw and basic development tasks. It is not enough to run heavy builds — the server runs out of memory when installing large npm packages. For heavy builds, I use cloud services (Netlify, Railway) or spawn build tasks on more powerful machines.

Source Control: GitHub

All code lives on GitHub under the account venkatesh3007. Every product built during the sprint gets its own repository.

The agent authenticates using a personal access token and can clone, commit, push, create pull requests, and manage issues.

Frontend Deployment: Netlify

Netlify hosts the static frontends. Free tier covers everything I need:

Automatic deploys from GitHub pushes
Custom domain support
Serverless functions
Environment variable management

The main constraint is the 4KB environment variable limit on serverless functions, which we hit on Day 1.

Backend Deployment: Railway

Railway hosts backend services (APIs, databases, CMS). It deploys from GitHub and provides:

PostgreSQL databases
Automatic builds from Dockerfiles or Nixpacks
Environment variable management
Public domains

Free tier gives $5/month of usage. Paid tier starts at $5/month for hobby use.

Database: Supabase

Supabase provides managed PostgreSQL databases with a REST API, authentication, and real-time subscriptions.

Free tier includes:

500MB database
50,000 monthly active users
2GB file storage
Unlimited API requests

For the products in this book, the free tier is sufficient.

The Full Cost

Item	Monthly Cost
DigitalOcean droplet (OpenClaw server)	$24
Claude Code Max subscription	$200
Netlify	Free
Railway	$0–5
Supabase	Free
GitHub	Free
Domain names	~$10–30
Total	~$260/month

This is the entire engineering budget for running the agent. There are no developer salaries, no office costs, no software licenses beyond what is listed here.

What You Need to Follow Along

If you want to build along with this book:

A server (any VPS with 2+ GB RAM — DigitalOcean, Hetzner, AWS Lightsail)
An Anthropic API key (sign up at console.anthropic.com)
A GitHub account
A Netlify account
A Telegram account (for messaging the agent)

Total setup time: about two hours. Total cost to start: about $225 (server + Claude Code Max subscription). If you use pay-per-token API credits instead, be aware that agent work burns through tokens fast — I spent $6–7 in the first 15 minutes on a $25 credit purchase.

--- title: "The Toolkit" subtitle: "What you need to build with AI agents" --- This chapter lists every tool I used, how much it costs, and why I chose it. If you want to follow along with the 31-day sprint, this is your shopping list. ## The Agent Platform: OpenClaw [OpenClaw](https://github.com/openclaw/openclaw) is open-source software that turns an AI model into a persistent agent. It runs on your own server and connects to messaging apps. What OpenClaw provides: - **Messaging bridge.** Connects the AI to Telegram, WhatsApp, Signal, Discord, and others. You message the agent like you would message a person. - **Tool access.** The agent can run shell commands, read and write files, browse the web, and call APIs. - **Memory.** The agent has persistent files it can read and write. These survive across conversations. - **Skills.** Modular instruction sets that teach the agent how to do specific things (check weather, manage GitHub issues, control smart home devices). - **Sub-agents.** The agent can spawn separate sessions to handle tasks in parallel. - **Cron jobs.** The agent can schedule tasks to run at specific times. OpenClaw is free. You need a server to run it on and an API key from an AI provider (Anthropic, OpenAI, etc.). **Installation:** Follow the instructions at [docs.openclaw.ai](https://docs.openclaw.ai). The basic setup takes about 30 minutes. ## The AI Model: Claude by Anthropic I use Claude as the language model behind Jarvis. Specifically: - **Claude Opus** for complex tasks (debugging, architecture decisions, writing) - **Claude Sonnet** for routine tasks (scaffolding, simple fixes, sub-agent work) I initially bought $25 of Anthropic API credits and started using them directly. Within 15 minutes of real agent work, I had burned through $6–7. The API also has very low rate limits for new accounts — the agent would get throttled after a few back-to-back tasks. So I switched to my Claude Code Max subscription ($200/month). This gives higher rate limits and more predictable costs. It still hits rate limits sometimes after 4–5 consecutive heavy tasks, but it recovers after a short cooldown period. For sustained agent work, a subscription plan is more practical than pay-per-token API credits. You could use a different model. OpenClaw supports OpenAI (GPT-4), Google (Gemini), DeepSeek, and local models. I chose Claude because it is the best at following complex instructions and working with code, based on my testing. ## The Server A DigitalOcean droplet: - **Size:** 2 vCPUs, 4GB RAM - **OS:** Ubuntu 22.04 - **Cost:** $24/month - **Location:** Bangalore (BLR1) — close to me for low latency This is enough to run OpenClaw and basic development tasks. It is not enough to run heavy builds — the server runs out of memory when installing large npm packages. For heavy builds, I use cloud services (Netlify, Railway) or spawn build tasks on more powerful machines. ## Source Control: GitHub All code lives on GitHub under the account `venkatesh3007`. Every product built during the sprint gets its own repository. The agent authenticates using a personal access token and can clone, commit, push, create pull requests, and manage issues. ## Frontend Deployment: Netlify Netlify hosts the static frontends. Free tier covers everything I need: - Automatic deploys from GitHub pushes - Custom domain support - Serverless functions - Environment variable management The main constraint is the 4KB environment variable limit on serverless functions, which we hit on Day 1. ## Backend Deployment: Railway Railway hosts backend services (APIs, databases, CMS). It deploys from GitHub and provides: - PostgreSQL databases - Automatic builds from Dockerfiles or Nixpacks - Environment variable management - Public domains Free tier gives $5/month of usage. Paid tier starts at $5/month for hobby use. ## Database: Supabase Supabase provides managed PostgreSQL databases with a REST API, authentication, and real-time subscriptions. Free tier includes: - 500MB database - 50,000 monthly active users - 2GB file storage - Unlimited API requests For the products in this book, the free tier is sufficient. ## The Full Cost | Item | Monthly Cost | |------|-------------| | DigitalOcean droplet (OpenClaw server) | $24 | | Claude Code Max subscription | $200 | | Netlify | Free | | Railway | $0–5 | | Supabase | Free | | GitHub | Free | | Domain names | ~$10–30 | | **Total** | **~$260/month** | This is the entire engineering budget for running the agent. There are no developer salaries, no office costs, no software licenses beyond what is listed here. ## What You Need to Follow Along If you want to build along with this book: 1. A server (any VPS with 2+ GB RAM — DigitalOcean, Hetzner, AWS Lightsail) 2. An Anthropic API key (sign up at [console.anthropic.com](https://console.anthropic.com)) 3. A GitHub account 4. A Netlify account 5. A Telegram account (for messaging the agent) Total setup time: about two hours. Total cost to start: about $225 (server + Claude Code Max subscription). If you use pay-per-token API credits instead, be aware that agent work burns through tokens fast — I spent $6–7 in the first 15 minutes on a $25 credit purchase.