MiroFish Goes Fully Local, Liberati is Live and a Lifetime Giveaway!

My first MiroFish video blew up! It passed 13,000 views and 88 comments, with heaps of great feedback, so I owed you all an update. In this video I launch Liberati, my new community for people building with decentralised AI at home, show the big upgrades to MiroFish Research since last week, and finish with a giveaway for lifetime access.

What is Liberati?

I've wanted to set up proper support for people building AI at home for a while, and now it exists at liberati.life. It's a Skool community for people who want to take the power back from big tech: no huge monthly API subscriptions, models running locally, and your data staying on your own hardware instead of going out to a middleman.

I'm honestly not chasing the "do 10x more work" productivity hype. What I care about is staying on top of the latest AI tools, actually testing them, getting them running with Ollama and local models, and making them useful in your life. Ultimately I see AI freeing people up to be more creative and do what they actually want to do.

How it works right now:

• Pop your name and email into the newsletter form at the bottom of the site and it redirects you to Skool
• There are three tiers, and the first one is literally the price of a coffee, $4 a month
• That gets you the code templates I build, access to the GitHub repos and a community forum
• Weekly community sessions are coming, and eventually one-on-one support for those who want it

I want to keep it affordable for everyone, whether you're in Australia, the Philippines or the USA, and just need some help getting the latest tech running on your own computer.

What's new in MiroFish Research?

Last week I promised you could run MiroFish without paying for anything, and the code is now updated. No cloud memory subscriptions and no paid API access for inference. Memory runs locally on Postgres with pgvector, and Ollama hosts the AI models.

The dashboard has had a lot of work too. It pulls in news from a bunch of feeds (some through MCP servers, some through RSS) and turns it into a format the AI can read. In the demo there are 107 articles coming in, filtered by topic, with oil and stocks available so far. Each topic can be shown four ways: a global map of where the news is coming from, a chart for price feeds, a table, and a plain article view with images. It also pulls historical news, going back years, so the AI has context like what happened to oil during previous Middle East wars.

From the current and historical news, the AI suggests possible future events. I picked the most topical one, "Iran war escalates to full-scale invasion", and hit Run Analysis. Four AI personas (like an oil market trader and a consumer impact economist) each analyse it, then debate each other's answers over a few rounds before writing a consensus report.

The result was a bit scary. The consensus came out at $364 a barrel, mostly because the economist persona went to $1,200 while the oil trader said $125. I don't think that's right, it would be absolutely crazy for the world economy! But that's the point of a swarm. You could throw another 10 agents with different perspectives at it and see where it lands. Also worth noting, this ran on a tiny Qwen 3.5 model, so bump it up to the 4B or 9B for better results.

Why am I building this?

I don't just build things, I actually use them. MiroFish Research is my personal demo and it's going to power a new YouTube channel about macro and geopolitics based on real data feeds that I try to keep non-biased. I'm not a US citizen or a Chinese citizen, I sit in the middle and just want to look at the data. Subscribe here: The MacroLedger.

If you missed the first one, catch it here: my first MiroFish video.

What's coming next?

Next week's template is Paperclip with several OpenClaw agents running, basically a startup in a box where you assign different roles to different agents. After that I'm testing ClawTeam with nanobot, both from the same developer, HKUDS.

How do you win lifetime access?

Drop the funniest comment you can on the video and the best one wins lifetime access to Liberati. Keep it within YouTube's terms please 😉

All the source code is in the community at liberati.life. See you in the next one, cheers!