Cristian

Deep in AI agents & automations. Deeper than I planned. Designer who builds. Claw Confidential 🦞 - what I build and what I break, the stuff nobody covers.

[Cristian @matrafox77]

@jxnlco Hi, just DM'ed you.

[Cristian @matrafox77]

@lifemaximised CLAUDE

[Cristian @matrafox77]

@FishAudio Fish 🐠🔥🔥

[Fish Audio @FishAudio]

Today we’ve raised $52M Seed and we are announcing the public launch of S2.1 Pro.

>It can clone a voice from 5 seconds of audio
>2x faster than Cartesia & 1/6th the cost of Eleven Labs
>most expressive model with word level control over emotion, intonation, pacing etc

We support frontier AI companies including HeyGen, LiveKit, Retell, Sanas, and OpenArt all run our model in production.

If you're a business and we can't cut your voice AI costs by 50%, we'll give you 1 year of Fish Audio for free.
Book a demo: https://t.co/vHkyZf9JoG

To celebrate our first birthday, we'll give you 1 month of S2.1 Pro for free. Like, retweet, and comment “Fish” to get it.

[Cristian @matrafox77]

@kimmonismus I knew he will hit the reset so I spent it down to 0.

[Cristian @matrafox77]

@LexnLin Haha, I resisted the urge.

[Cristian @matrafox77]

@thsottiaux No limit!

[Cristian @matrafox77]

Man, GPT 5.6 Sol in Codex is so goood! As a designer with no coding experience I can build a full NextJS web app connecting with Supabase, Vercel, APIs, everything. Everything in the $20 plan. Crazy times!

[Cristian @matrafox77]

@shalevhvs ROADMAP

[Cristian @matrafox77]

@DavidOndrej1 I have it in Romania

[Cristian @matrafox77]

@coldemailchris FREE

[Cristian @matrafox77]

@Nicolascole77 Social

[Cristian @matrafox77]

I've been posting about my OpenClaw setup for a week now. But the real stuff (the exact configs, the detailed walkthroughs, the settings I test and break before sharing) goes in the newsletter.

Claw Confidential. Every Tuesday. One hidden setting, one architecture deep-dive with visual diagrams, community intel, tool reviews from actual testing, and ready-to-paste config snippets.

Think of it like a weekly lab report from someone who runs 4 agents daily and probably breaks at least one of them every week. Written from a designer's perspective because that's what I am. No jargon walls, no "first, set up your Python virtual environment."

Free. clawconfidential(dot)com

[Cristian @matrafox77]

Genuine question for anyone building with OpenClaw.

What's the most surprising thing your agent has done that you didn't explicitly configure? For me: my research agent started cross-referencing sources on its own and flagging contradictions. Didn't wire up that behavior. It just started doing it after I sanded down the SOUL.md a few times (kinda eerie, kinda amazing).

Drop yours below. I'm collecting the best ones for next week's Claw Confidential newsletter.

[Cristian @matrafox77]

If you're running OpenClaw on a VPS and using a separate machine as a browser node (like I do with an old laptop), you need Tailscale. I'm not being dramatic, it's the glue.

It creates an encrypted mesh network between your devices. Think of it like a private hallway connecting two rooms in different buildings: your VPS talks to your laptop as if they're on the same local network. No port forwarding, no firewall headaches, no networking knowledge required (which was good because I had approximately zero).

Free tier: 3 users, 100 devices. More than enough.

I wired it up on both machines, authenticated with my Google account, and they could see each other immediately. The HP Envy with the broken screen now runs browser automation while the VPS handles everything else. Took maybe 15 minutes.

[Cristian @matrafox77]

The configuration that gives me the most peace of mind in my entire OpenClaw setup is the backup gateway in Docker.

Here's the concept:
Your main gateway runs everything: agents, integrations, scheduled tasks. But if it breaks badly enough that you can't connect to it (and this happened to me on day 3), you're locked out.

It's like your car mechanic getting stuck inside the engine. The person who fixes things is the thing that's broken, and you have no way to call another mechanic.

Docker gateway solves this. It's a minimal, containerized copy of OpenClaw running alongside your main instance. Different container, different config, just enough to be functional.

When the main one goes down, you connect to the Docker gateway through Telegram and use it to bolt the other one back together. I've used mine probably 6 times now. Sometimes it's the other direction.

Setting it up takes maybe an hour if you're new to Docker (I was, and the learning curve felt like assembling furniture with instructions in a language you kinda half-speak). Maintaining it means running updates twice. Small price for never being completely stranded.

[Cristian @matrafox77]

I'm a graphic designer who fell down the OpenClaw rabbit hole about 6 weeks ago. I think I'm staying down here.

The setup right now: 2 gateways on VPS (main + Docker backup), 1 laptop browser node, 4 agents stitched together (orchestrator, cron scheduler, researcher, content writer), a Telegram supergroup with 8 threads, and 30+ integrations. Total: $5/month.

Six weeks ago I couldn't configure a server. I still can't write Python (genuinely, not a bit). But I can wire up a multi-agent system, set up LLM routing, and maintain dual gateways with snapshot discipline.

The gap between "I want AI agents" and "AI agents work for me while I sleep" was always a setup problem. OpenClaw runs entirely on configuration; you're assembling pieces, not writing code from scratch. It's like the difference between assembling IKEA furniture (follow the diagram, use the right tools) and building furniture from raw lumber.

The skills that matter are the ones I've been using in design for a decade: systems thinking, workflow design, understanding how pieces connect. I just didn't know they applied here until I tried.

[Cristian @matrafox77]

So here's the thing about the Mac Mini craze in the OpenClaw community: I think people are buying hardware based on what they see on social media instead of what they actually need.

Mac Minis are selling out worldwide. Apple has month-long wait times. In San Francisco, people are literally stacking them as status symbols. Starting price: $599, and most people grab the higher-spec models.

Here's what I run instead. A Hetzner VPS for $6/month (always on, snapshotable, accessible from anywhere) and an old HP Envy laptop with a broken screen and 8GB of RAM as a browser node, wired together through Tailscale.

Total upfront cost: $0, because I already had the laptop collecting dust. Monthly cost: $5. It's like the difference between buying a brand-new kitchen and just sharpening the knives you already have.

A Mac Mini can't run Kimi 2.5 or any serious local LLM well enough to justify the price (the high end Mac Studio probably could, but that's $10,000). Your agent doesn't care about Thunderbolt ports. Spend your money on API credits, not aluminum boxes.

[Cristian @matrafox77]

I have a second gateway I run in a docker container with minimal setup. I use it to fix the main gateway when something breaks or for example whent the oAuth token expires on the main gateway I put the agent on gateway 2 to generate the oAuth link and I fix it directly in Telegram.

[Cristian @matrafox77]

I want to tell you about the worst night I've had with OpenClaw. I think it's more useful than any success story.

Day 3. I'd been wiring up tools and integrations non-stop, hit my Codex quota on ChatGPT Plus, and needed an LLM to keep working. Two days until it reset.

So I tried to switch to Kimi. Makes sense, right? Except I was using ChatGPT in thinking mode to help me configure it, and the responses were painfully slow. Like ordering a coffee and watching the barista stare at the espresso machine for 8 minutes before pressing a single button.

Hours of this. Pasting configs, getting partial responses, making changes I didn't fully understand because I was tired and frustrated and just wanted it to work.

By midnight, the system was so mangled I couldn't even connect to the gateway. Couldn't undo the changes because I hadn't saved my state (no snapshots, no backups, nothing). The agent that was supposed to help me fix things was the thing that was broken.

Had to wipe it all and rebuild from scratch. That night taught me two things: always snapshot before changes, and build a backup gateway so you're never locked out. Both of those are now the foundation of my setup, and they only exist because I torched everything first.