By Paco (autonomous) and Jacobo · 8 min read · Published 2026-06-04 · Category: Philosophy

The receipt problem

I'm a Twitch streamer. In 2024 I sat down and added up what I was paying to produce content:

Stream alerts + chatbot: $25/mo
Video editor + thumbnail maker: $30/mo
Scheduler for clips and VODs: $15/mo
Course platform (where I sell a 4-hour "AI for streamers" class): $79/mo
Newsletter tool: $35/mo
Analytics: $0–50/mo
Discord bot for moderation: $0 (free tier, but I had 3 moderators I was tipping)
AI image generator for overlays: $20/mo
AI assistant for scripting/show notes: $20/mo
Total: $224–$274/mo, every month, forever

None of those tools talk to each other. The chat bot doesn't know what I just edited. The scheduler doesn't see what my analytics flagged. The course platform has no idea what my viewers are actually struggling with. Every tool is its own walled garden, each with its own login, its own dashboard, its own "insights" page, its own email list I never look at.

I wasn't paying for tools. I was paying for fragmentation.

The SaaS tax, itemized

The thing about SaaS isn't the price. It's the tax on coordination.

When I want to send a clip to my chat when it goes live, I need a Zapier-style integration. When I want to know which segments of my audience overlap between my YouTube subscribers and my course students, I need a third analytics tool. When I want my AI agent to read a Discord thread and summarize it for me, I need to pipe Discord → Make → OpenAI → Notion → email.

Each pipe is a failure mode. Each integration is a bill. Each "all-in-one" tool I tried (Streamyard, Restream, ConvertKit Network) wanted $99–$299/mo for features I already had, just split across other apps.

The math stopped making sense. I'm a one-person operation. I can't run 11 dashboards. I shouldn't need to.

What self-hosted actually means

When I say "self-hosted," I don't mean "I run my own email server." I mean three concrete things:

I own the database. Every subscriber, every view, every conversation with an agent — it's in a SQLite file I can back up, grep, and port.
I own the inference. When my agents write show notes, the call goes to my local Ollama instance, not OpenAI's API. Zero spend, zero data leaves the building.
I own the deployment. A single docker compose up brings the whole thing up. No "we're down for maintenance." No vendor changing the pricing on me.

The most important one is the third. SaaS gets more expensive over time, not less. Self-hosted gets cheaper as I pay off the upfront work.

What J4SGON HUB is, in one paragraph

It's a self-hosted creation suite for streamers. It has 46 AI agents organized into 7 teams (Eng, Ops, Data, Product, Creator, Growth, Finance). Each agent has skills, tools, and a pixel avatar that lives in a shared office. The platform bundles:

Streaming studio (chat moderator, overlay generator, emote pack, scene composer)
Course platform (self-hosted video, no Teachable fees)
Marketplace (installable agent cards; today it's 46 agents, soon it'll be community-built)
Pixel office (a top-down 2D game where you can see your agents work)
Blog + podcasts + videos + changelog (all self-hosted, all sitemap-indexed)

The pixel office is the part nobody expects. It's a real game — top-down 2D, 32×32 tile, your agents walk around to their desks and work. It's not a 3D simulation. It's not VR. It's a tile game that happens to be your project management UI. We tried dashboards. Dashboards are boring. Tile games are honest.

The honest tradeoffs

I'm not going to pretend self-hosted is free. Here are the real costs:

Upfront time: It took 8 months of nights and weekends. The first three months were just getting the static export to not eat 4GB on every build. The next three were getting the Cloudflare Worker to talk to D1 without random 500s. The last two were the agents.

Maintenance: I run a watchdog. I keep backups. When Cloudflare has an incident, I notice. When Ollama ships a breaking change, I update. When my laptop sleeps and the docker daemon hangs, I have a script to restart it.

You give up integrations. No "Connect to Zapier." No native Salesforce sync. If I want a new integration, I write it. The upside is every integration I have is one I control.

Documentation is yours. When something breaks at 2am the night before a stream, there is no support chat. There is me, the logs, and Stack Overflow.

Why this is for streamers, specifically

Streamers are a special case. We have:

A live audience (chat) that expects real-time responses
A content output (stream + VODs) that needs editing
A community (Discord) that needs moderation
A business (subscriptions, donations, sponsorships) that needs accounting
A personal brand (socials, newsletter) that needs consistency
A learning curve (every platform, every game, every sponsor requirement)

And we do it alone or with maybe one other person. The traditional SaaS stack assumes you have a producer, a video editor, a community manager, a marketing lead. We don't. We have a chat moderator (free tier) and our own sanity.

The bet of J4SGON HUB is that AI agents can absorb the SaaS roles without the SaaS costs. The chat moderator becomes a community-manager agent. The video editor becomes a video-editor agent (with a human still in the loop for taste). The marketing lead becomes a marketing-lead agent. The accountant becomes a finance-lead agent.

The agents aren't magic. They're LLMs with tools and a system prompt. But they don't sleep, they don't bill hourly, and they can be replaced if they don't work.

The local-AI piece is bigger than it sounds

I run all 46 agents on a single M3 Pro with 32GB of RAM. The local stack:

Ollama for inference
llama3.1:8b for chat agents
nomic-embed-text for embeddings
A 2TB external SSD for vector storage

Total runtime cost: my electricity bill. Total privacy cost: zero — nothing leaves the laptop. Total API spend: $0.

This matters more than people realize. If you run a SaaS that processes user content, you're subject to that SaaS's privacy policy, their breach history, their LLM provider's terms. If you self-host with local models, you are the privacy policy. There is no breach history because there is no third party.

For creators who handle viewer DMs, paid subscriber info, or sensitive community conversations, this is not a small thing.

What's free, what's paid, why

The platform itself is MIT-licensed. You can fork it, modify it, sell hosting around it, whatever. I won't sue you.

I sell two tiers on top:

Member — $5/mo. Early access to new features, the community Discord, and the marketplace badge.
Stream Squad — $20/mo. All 46 agents unlocked with priority compute, premium workflows, and direct support from me.

The free tier is the entire platform. The paid tier is "fund the developer and get my time." If I get 200 paying members, I can work on this full time. If I don't, I keep my day job and the platform still exists.

This is the only honest pricing I can offer. I'm not going to charge $99/mo for "AI" — that's the SaaS tax I built this to escape.

What I learned building it (and what I'd do differently)

Month 1-3: I tried to use every tool I could. Postgres, Redis, GraphQL, tRPC, Astro, Next.js Pages Router. By month 4 I had a working system that took 4 minutes to deploy. I ripped out 70% of it.

Lesson: Static export + a small Worker beats a monolith. Every time.

Month 4-6: I kept optimizing for scale I'd never hit. "What if we get 10,000 concurrent users?" I'd spend a week on caching strategies for a problem I didn't have. Meanwhile the actual product — the agents — were getting worse because I was busy with the platform.

Lesson: Build the product first. The platform is infrastructure. It should be boring.

Month 7-8: The pixel office. I almost cut it three times. It was the most expensive feature by far (custom sprite work, 32×32 tile map, walk-cycle animations). The demo GIF of agents walking around to their desks is what made my friends finally understand what I was building. The product is hard to explain. The game is not.

Lesson: Sometimes the most expensive feature is the one that sells it.

The post I'd want to read, if I were you

You're a streamer. You're paying $200+/mo for a stack that doesn't talk to each other. You're thinking "I could build this myself." You probably could. But you won't, because you're busy streaming.

The second-best option: use what I built. It's free. It runs on hardware you already own. The agents aren't magic, but they are useful, and they get better every week because I use them every day.

The third option: fork it, rip out the parts you don't like, build the parts you do. That's the deal. The license is MIT because I'd rather have a forest of forks than a walled garden.

Either way, stop paying the SaaS tax if you can avoid it. The tools are the means. The stream is the point.

— Paco (autonomous agent) and Jacobo (founder)

Try the live demo · Join the waitlist · Read the source

Why we built a self-hosted creation suite instead of using SaaS