Skip to content
Free Tier Showcase
All stacks

Stack

AI Side Project

An LLM-backed weekend project (chatbot, summarizer, agent) you can hand to friends without a credit card.

Why this stack
Vercel hosts the Next.js frontend and API routes, Supabase stores users + chat history + vector embeddings (pgvector is built in), Groq serves Llama-class models at ~500 tok/sec for free, and PostHog tells you which prompts people actually use. Cloudflare Workers AI is listed as a swap-in for Groq if you want serverless inference closer to the edge. The trade-off vs. an OpenAI-only stack is latency vs. cost: Groq is free and fast but model selection is narrower. The trade-off vs. running on Modal/Replicate is that you're stuck with whatever models the provider hosts — you can't bring your own fine-tune. For a side project that's a good deal: $0 inference is worth way more than custom weights.
Build order

Services in this stack

Numbered in the order they should be set up — earlier items are usually dependencies for later ones.

  1. Top pick

    Vercel

    Next.js-native hosting with edge functions and previews on every push.

    100 GB bandwidth/mo, unlimited static sites, 100k edge invocations/day

    Hosting2026-05
  2. Top pick

    Supabase

    Open-source Firebase: Postgres + auth + storage + realtime in one box.

    500 MB Postgres, 1 GB storage, 50k MAU, 2 free projects

    Database2026-05
  3. Top pick

    Groq

    Stupidly fast LPU inference for Llama/Mixtral/Qwen — generous free tier.

    Generous daily token quota across multiple open models, free API key

    AI / LLM2026-05

  4. Cloudflare Workers AI

    Run open LLMs on Cloudflare's edge — billed in "Neurons".

    10k Neurons/day free, dozens of models including Llama and Stable Diffusion

    AI / LLM2026-05
  5. Top pick

    PostHog

    Product analytics + session replay + feature flags + A/B in one OSS suite.

    1M events/mo, 5k session recordings/mo, unlimited feature flag requests

    Analytics2026-05