404 Conference
Free Freaking AI
How to run your entire AI stack on $0.
Yan Gonzalez
AI moves fast.
The economics move faster.
Today — a recap of what's worth paying for, and what's free.
Four Areas of Free
You're Already Paying for It.
Each one ships with a coding agent.
The Sub Is Subsidized.
Same model. Same tokens. Wildly different bills.
💳 API direct (Opus)
$5 / $25
per 1M tokens, in / out. Full price.
✨ Claude Max $200
$2,700/moat Sonnet rates
~$8,100/moat Opus rates
For $200. Anthropic eats the difference.
🪣 Replit / Cursor
API + markup
Same API tokens — they charge extra for the wrapper.
Cheapest tokens on Earth = sub direct. Skip the middleman.
Your Chat Is a Train.
Every reply re-sends the whole conversation. Charged both ways. Every time.
With a sub, you don't see the train grow. On the API, you do.
Pick Your Coding Assistant
You probably have access to one. Match your sub. Or grab the free one.
- Parallel tasks, full repo
- Browser + terminal
- Skills, Subagents, MCP
- Audits, ships, ops
- Agent-first IDE, Gems
- 1M-token context
- Terminal TUI, any model
- $0 — yours forever
One of these is already yours. Go turn it on.
Same Chat. Now With Hands.
Real things Claude does for me every day. Not hypothetical.
SSH into my servers
Restart stuck services, tail logs, patch nginx — all from one prompt.
Drive my browser
Log in, fill forms, scrape dashboards I can't API into. Click buttons for me.
Ship to Cloudflare Pages
Edit copy on client sites, commit, deploy. I review the diff after.
If you only use it to write copy, you're using 10% of what you already own.
Don't Waste the Flagship Models
For 80% of tasks, the mini model is cheaper and already in your sub.
GPT-5.3 Codex Spark
ChatGPT Pro · GPT-5.4-mini on Plus / Business
Quick edits, Q&A, low-stakes rewrites.
Claude Haiku · Sonnet
Claude Pro / Max
Haiku for quick stuff. Sonnet for real work. Save Opus for hard problems.
Gemini Flash
Gemini Sub / Antigravity
Drafts, bulk processing, repetitive tasks.
02 · APIs that are actually free
Google AI Studio (Gold Mine)
No credit card. No expiry. Here's what the numbers actually mean.
Deep research, client audits, complex rewrites. Use it for hard problems.
Bulk meta titles, content drafts, daily automation. Your workhorse.
Frontier quality, open weights. Run it in AI Studio or self-host it.
10× more daily requests than 31B. Use it for volume work and repetitive tasks.
Four frontier models. $0. No card. No expiry. aistudio.google.com
More Free Providers. Real Models.
Beyond Google. Three more $0 menus, sign up today.
⚡ Cerebras Cloud
GLM 5.1 · Qwen 3.5 397B
1M tokens/day free. Fastest inference available. No cold starts.
🚀 Groq
Kimi K2 · DeepSeek V3
Free API key, instant signup. 300+ tok/sec on LPU hardware.
🖥️ Ollama
Qwen3-Coder 480B
Run locally or on Ollama Cloud. Zero per-token cost once it's running.
Stack any three with AI Studio and you've replaced 90% of paid API calls.
Edge Gallery
Real AI models on your phone. Offline. $0. Nothing leaves the device.
No API keys. No bill.
Private. On-device only.
Gemma · Llama · Phi — client demos with zero connectivity, privacy-sensitive work
Point your camera. Download it before the next slide.
Local Is "Free" If You Own the Metal
Zero cost per token — after the hardware bill clears.
🖥️ Mac Mini M4 Pro
$999
- 32GB unified memory
- Runs Qwen / Llama 8B–14B comfortably
- Break-even vs $20/mo MiniMax 2.7: ~4.2 years
🖥️ Mac Studio M4 Max
$5,999
- 256GB unified memory
- Runs 70B+ models natively
- Break-even vs $200/mo Claude Max: ~2.5 years
If you don't need offline, privacy, or 24/7 throughput — the sub is cheaper. Do the math before you buy the metal.
Remote Access. Free.
Reach your machines, reach your clients' machines. No per-seat bill.
Tailscale
100 devices · 3 users
- Zero-config mesh VPN on WireGuard
- SSH into anything, from anywhere
- Personal plan, free forever
RustDesk
Open source · Self-host
- TeamViewer / AnyDesk replacement
- Run your own server or use theirs
- AGPL, no per-seat license
Stop paying $30/user/month for what these do free.
Data APIs. Free Tier.
The APIs your agents need to be useful. No credit card to start.
Brave Search API
$5/mo credit≈ 1,000 queries/month
- Independent search index, not Google/Bing
- Ground your agents in real web results
- 1 query/sec on free tier
Supadata
100 credits/mo≈ 100 YouTube transcripts
- YouTube transcription by URL
- Feed videos into your LLM pipeline
- Client research, content repurposing
Search + transcription — the two most common agent superpowers, free to start.
GitHub Is a Candy Store.
Think App Store / Google Play — but for your AI agents. Free.
MCP Servers
Plug-in skillsfor Claude, Cursor, and friends
- Browser control, Playwright, Puppeteer
- Databases, filesystems, APIs
- Thousands of servers on GitHub
Open Source
Ready-made toolsyour agent clones and runs
- FFmpeg, yt-dlp, ImageMagick, pandoc
- Whisper, Ollama, LibreOffice
- If a SaaS does it, GitHub has it free
Your agent doesn't need a subscription. It needs a git clone.
Free Code. Free Hosting.
A live URL costs $0. Most people still pay for one.
🐙 GitHub
- Unlimited public & private repos
- Free CI/CD with Actions
- Pages hosting included
🌩️ Cloudflare Pages
- Unlimited static sites, $0
- Global CDN, free SSL
- Auto-deploy on git push
Your agent writes the code. These two host it. Watch.
Try It · 15 minutes
Let's Ship a Site.
Kick off the build. The next slide is a song. Your URL goes live before it ends.
- Gmail account
- Claude Code (or OpenCode)
- 10 min of focus
- Sign up: GitHub + Cloudflare (Gmail one-click)
- Open Claude Code in an empty folder
- Prompt: "Build a hello-world landing page. Push to a new GitHub repo. Set up Cloudflare Pages to deploy it."
- Paste the GitHub + Cloudflare auth links it asks for
- Open the live URL. Show your neighbor.
Hit go now. Music plays next while it builds.
A gift · Claude Code skill
site-to-astro. Yours.
Point it at any site. Get a blazing-fast Astro clone. Free.
github.com/truewebmaster77/
site-to-astro
- WordPress, static HTML, legacy CMS → Astro
- Tailwind v4, i18n, Cloudflare Pages
- Perfect PageSpeed out of the box
Scan it. Clone it. Ship a client site by Monday.
While your build runs
404: That's Me
By the time the song ends, your URL is live. Welcome to the internet.
All four areas. One config. Bills stop moving.
Open source. Free. Shipping to OpenClaw today.
Real Savings
Actual /ffai-stats output from my own deployment.
That's me, casually. Playing with my OpenClaw and running a bunch of tests.
Demo Time
Let's actually run it.
Before You Close This Deck.
Turn on. Sign up. Stack.
your coding agent — you already pay for it.
for Google AI Studio — free, no card.
your OpenClaw with FFAI when you're ready for the rest.
Do one today. One this week. That's it.
