Thursday, November 19, 2026
HomeEntrepreneurI Constructed an AI Chatbot for $0. Here is the Actual Stack.

I Constructed an AI Chatbot for $0. Here is the Actual Stack.


Opinions expressed by Entrepreneur contributors are their very own.

Key Takeaways

  • Each layer of a contemporary chatbot, from the brand to the database to the voice output, now has a real free tier, not only a trial.
  • Virtually all of those instruments cap utilization somewhat than time. Which means the restrict you’ll hit first is quantity of customers or requests, not a countdown clock.
  • The one layer value watching intently is the coding assistant itself. Claude Code’s free chat plan doesn’t embrace it, so sustained use ultimately means both a Visitor Go, or leaning extra on Codex’s personal free plan.

I’m an AI engineer and I’m at present constructing ChatMakr, an AI-powered platform that creates multi-agent chatbots in seconds. I additionally constructed Rune Rating, a sports activities analytics chatbot, and it price me $0.

From the surface, it seems to be like a easy chatbot. Underneath the hood, it’s a distributed system that has to deal with the freshness of sports activities information, person rephrasing, extraction failures, LLM name prices, streaming, reminiscence, fallbacks and safety. Right here’s the stack I truly used to construct it layer by layer, what I did with every software and what’s free about it.

Branding: Figma

Earlier than writing a line of code, a chatbot wants a face. I used Figma to generate the chatbot’s emblem straight from a textual content immediate describing the type and temper I needed, then iterated on it just a few instances inside Figma itself somewhat than handing it to a designer. Figma’s personal free plan covers particular person use with limitless file storage.

Code internet hosting: GitHub

The entire challenge lives in a GitHub repository, non-public whereas the chatbot was nonetheless tough, and it’s the identical repo each Claude and Codex hook up with. GitHub’s free plan offers limitless non-public repositories with a restricted characteristic set and a couple of,000 minutes a month of Actions included on GitHub Free for normal runners on non-public repos, sufficient to run assessments and fundamental automation with out hitting a wall. Commonplace runners on public repositories keep free regardless of how a lot you employ them.

Growth: Claude Code, with Codex because the decide

Claude Code did the precise constructing, the chatbot’s backend and the AI brokers working inside it. I stored it linked on to the GitHub repository and by no means let it contact the principle department straight: each process ran by itself department, and nothing merged till I’d reviewed the diff myself and authorized the pull request. The real free route in is a Visitor Go: if somebody already pays for Claude Max, they will ship you one, and it unlocks a full week of free Professional-level entry.

I gave Codex a distinct job, linked to the identical repository however by no means writing manufacturing code itself. It reviewed what Claude had truly shipped, performing as an impartial decide on pull requests, and I additionally used it to draft the subsequent improvement immediate I’d hand again to Claude, successfully having one mannequin plan the work and one other execute it. Codex has a real free plan at $0 a month. Free accounts run on GPT-5.6 Terra, OpenAI’s cost-efficient mannequin for that tier. The free plan is capped at a utilization restrict somewhat than a time restrict, which is precisely the dimensions of job a reviewer function wants.

Area and safety: Cloudflare

I purchased the chatbot’s area title by way of Cloudflare, the place registration is often low-cost since Cloudflare sells at price with no markup, and stored the DNS in the identical place somewhat than splitting it throughout suppliers. From there I used Cloudflare’s free safety layer to guard the area, together with charge limiting guidelines to cease any single supply from hammering the chatbot’s endpoints. Cloudflare’s WAF documentation covers precisely this sort of rule, and the core protections, DNS, DDoS safety, and fundamental security measures, are included on the free plan.

Deployment: Railway

I used Railway for each the backend and the frontend, with a separate take a look at setting so I might ship adjustments shortly with out risking the stay model. New accounts get a one-time free credit score grant earlier than any card is required, and in contrast to some free hosts, Railway doesn’t put your service to sleep by default the best way its personal documentation describes for its optionally available Serverless characteristic, so there’s no chilly begin tax on the free tier except you flip that conduct on your self.

Authentication: Clerk

Customers join and log in by way of Clerk, and I constructed the entire stream utilizing Clerk’s personal take a look at setting first, so I might strive edge instances with out creating actual accounts or sending actual emails. The free plan covers as much as 50,000 month-to-month lively customers for free of charge, which is greater than sufficient room to maneuver from testing into an actual launch on the identical setup.

Database: Neon

Conversations, customers and utility information all sit in Neon, a serverless Postgres database that scales all the way down to zero and stops costing something the second no one’s utilizing it. The free tier contains sufficient storage and month-to-month compute hours for an actual early product, with no bank card required to start out.

Observability: Langfuse

As soon as the chatbot’s AI brokers have been truly working, I wanted to see what they have been doing below the hood: which prompts fired, the place they failed, and what they price. Langfuse traces all of that, and its free Pastime plan features a wholesome month-to-month allowance of tracing occasions, sufficient to debug and monitor an actual chatbot with out paying something.

Internet analysis: Exa and Firecrawl

I wired Exa and Firecrawl straight into the AI brokers chargeable for analysis contained in the chatbot’s runtime, so the chatbot itself can resolve when it must look one thing up. Exa handles the search, discovering related pages for a given query, and its free tier fingers new accounts beginning credit score plus a recurring month-to-month allowance with no fee technique required. It additionally ships a playground the place you may take a look at a search question straight within the browser earlier than wiring it into your agent. Firecrawl handles what comes subsequent, turning a web page Exa discovered into clear textual content the agent can truly learn, and its personal free plan features a month-to-month credit score allowance with no card wanted both.

Testing AI fashions: NVIDIA

Earlier than committing to a particular mannequin for the chatbot’s brokers, I used NVIDIA’s API Catalog to strive a number of with out paying per token. NVIDIA offers any developer free credit to check a variety of hosted fashions by way of its personal endpoints, which made it simple to match choices earlier than locking in whichever one truly powers the chatbot in manufacturing.

Voice: Azure AI Speech

The chatbot talks again, not simply varieties, so I used Azure AI Speech for textual content to speech, letting the chatbot change between a feminine or male neural voice relying on the setup. The free tier features a actual month-to-month allowance of neural voice characters for free of charge, sufficient for real each day conversations earlier than you’d ever see a invoice.

Key Takeaways

  • Each layer of a contemporary chatbot, from the brand to the database to the voice output, now has a real free tier, not only a trial.
  • Virtually all of those instruments cap utilization somewhat than time. Which means the restrict you’ll hit first is quantity of customers or requests, not a countdown clock.
  • The one layer value watching intently is the coding assistant itself. Claude Code’s free chat plan doesn’t embrace it, so sustained use ultimately means both a Visitor Go, or leaning extra on Codex’s personal free plan.

I’m an AI engineer and I’m at present constructing ChatMakr, an AI-powered platform that creates multi-agent chatbots in seconds. I additionally constructed Rune Rating, a sports activities analytics chatbot, and it price me $0.

From the surface, it seems to be like a easy chatbot. Underneath the hood, it’s a distributed system that has to deal with the freshness of sports activities information, person rephrasing, extraction failures, LLM name prices, streaming, reminiscence, fallbacks and safety. Right here’s the stack I truly used to construct it layer by layer, what I did with every software and what’s free about it.

Branding: Figma

Earlier than writing a line of code, a chatbot wants a face. I used Figma to generate the chatbot’s emblem straight from a textual content immediate describing the type and temper I needed, then iterated on it just a few instances inside Figma itself somewhat than handing it to a designer. Figma’s personal free plan covers particular person use with limitless file storage.

RELATED ARTICLES

Most Popular

Recent Comments