Gregory Luneau

Father, Entrepreneur, Musician, Developer, Martial Artist, Life Hacker, Builder, Sci-fi Reader

[Balboa Exchange @balboaexchange]

In Panama.

Built where two oceans meet, BalboaX serves qualified clients across Latin America and beyond. Anchored locally. Connected globally.

Discover more at https://t.co/EaUIG3f7e0.

[Balboa Exchange @balboaexchange]

High-touch OTC execution in bitcoin and major stablecoins, with settlement in USD, USDT or USDC. One relationship across digital assets and banking rails.

https://t.co/EaUIG3f7e0

[Balboa Exchange @balboaexchange]

Been a while. Almost 3.5 years, actually.

We weren't gone. We were getting ready.

Balboa Exchange is back on your timeline. More soon.

https://t.co/6vn3EuHriL

[Gregory Luneau @gluneau]

@roelandp For AI yes, but fear not, there is a skill for this!

[Peter Steinberger 🦞 @steipete]

OpenClaw deleted around 400k LOC of its own tests without much change in code coverage. Modern models just love writing tests for every tiny change, even if they aren't useful. This skill helped. https://t.co/a1jSBpfpns

[Sunovia Noir @SunoviaNoir]

Le loup, le renard, le lièvre - M'en revenant de Sainte-Hélène | Duo Folk Québécois

[Gregory Luneau @gluneau]

@mattyp Oh the irony! 🔥

[Gregory Luneau @gluneau]

@MatthewBerman What models can run on vps and how. To run skills and workflows.

[Gregory Luneau @gluneau]

@Adelgary @francispouliot_ Same, can’t see the media

[Gregory Luneau @gluneau]

@MyLatinLife https://t.co/vX1VQhpUYZ is what you’re looking for.

[Gregory Luneau @gluneau]

@francispouliot_ @Andrew_J_Howard @BullBitcoin_ Avec plaisir! 🫡

[Sunovia Noir @SunoviaNoir]

Un Canadien errant - Chanson Traditionnelle Québécoise | Hymne Patriotique Canadien-Français

[Gregory Luneau @gluneau]

@steipete Wow, this presentation is full of amazing insights. Everyone in there is worthy of a follow.

[Gregory Luneau @gluneau]

@francispouliot_ 👀

[FRANCIS ⚜️ BULLBITCOIN.COM @francispouliot_]

🚨 JOB OPENING AT BULL BITCOIN

We are hiring a senior backend developer. Share this with your friends. BULL is a phenomenal place to work with tight-knit team of highly competent engineers.

This is your chance to meaningfully contribute to The Mission!

https://t.co/V9rupH98Ix

[Gregory Luneau @gluneau]

I accidentally contributed to two open source projects. Here's how it happened.

Ollama Cloud has a free tier that's almost suspiciously generous. Kimi K2.5, GLM-5, MiniMax M2.5 -- big models, no credit card. I wanted to set them up as Discord bots through OpenClaw and just... poke at them. See what they're like.

They timed out. Every single message, at almost exactly 17 seconds.

I spent longer than I'd like to admit chasing it. The error said "LLM request timed out" so I went looking for timeout values in the source. Found a 30-second one. Patched it. Still 17 seconds. Found a 15-second one. Not it either. At some point I had patched three different compiled files and none of it mattered.

The real issue was embarrassingly mundane. I had `api: "openai-completions"` set on each model individually, which quietly overrode the provider-level `api: "ollama"` setting. So instead of hitting the native `/api/chat` endpoint, every request was going to `/v1/chat/completions`. That endpoint responds fine -- it just buries the model's thinking in a `reasoning` field and leaves `content` empty. OpenClaw read the empty content, retried a few times, gave up at 17 seconds.

Once I fixed that, I hit the actual bug. Ollama's native API uses `message.thinking` for thinking-capable models. OpenClaw's stream reader was checking `message.reasoning`. Different field name, never matched, content silently dropped.

I filed an issue. Dingding-leo had a complete fix posted in the comments within minutes -- both paths, streaming and non-streaming, ready to copy. Then ayanesakura turned it into a PR. Then vincentkoc cleaned it up and added backward compatibility for older Ollama versions that use `reasoning` instead. That version got merged. It's in 2026.3.7.

The other one started the same way. I use a CLI called `summarize` to pull YouTube transcripts. It kept failing on videos I knew should work. Took me a while to find it: failed transcript fetches were being cached with the same 30-day TTL as successful ones. So a single bad run would block retries for a month. I manually cleared the cache to move on, filed issue #114, and it was fixed in PR #115 -- failed results now get a 6-hour TTL instead.

Neither of these was planned. I just wanted free model access and kept hitting walls.

Filing the issues took maybe 10 minutes each. Describing what broke, what I tried, what I found. That's it. Other people did the rest.

@steipete thanks for all your great work.

---

If you want to see how I set up @ollama Cloud as Discord channel bots through @openclaw , leave a comment. Happy to put together a walkthrough if there's interest.

---

https://t.co/b18aoIGuVx https://t.co/SJtAmEs6fm

[Gregory Luneau @gluneau]

Are you using AI just for prompts, or for real system automation?

Vercel Analytics looks great. Clean dashboard, no setup. But there's no API. You can see the numbers, you just can't query them from code, or ask an AI agent to summarize them for you.

I run around 20 sites, a few apps, some half-baked experiments. @vercel was fine when I only cared about one or two of them. Once the portfolio got bigger, I wanted to ask "how are all my sites doing?" without opening 17 tabs.

So I asked my AI agent (I use @openclaw , a self-hosted agent that runs on my server) to figure something out. It confirmed Vercel had no analytics API, suggested Google Search Console and GA4, walked me through an OAuth approval, and got to work.

What it actually did:
It hit the @Namecheap API to pull the DNS records for all my domains. Got a unique verification token from Google for each one. Added the TXT records without touching anything else. Verified all of them with Search Console. Created GA4 properties, correct timezone, correct currency. Then pulled each GitHub repo, added the tracking snippet to the layout file, and opened a PR.

I merged the PRs.

The copy-paste version of that workflow would have cost me a couple of hours. The agent finished in about 15 minutes. The only thing it needed from me was OAuth approval at the start and a merge click at the end.

Now when I want to know how a site is doing, I just ask. The agent calls the GA4 Data API and tells me. No dashboards.

[LTX.io @ltx_io]

🚨 800 FREE CREDITS — 24 HOURS ONLY

We just launched LTX-2.3, our most powerful video model yet. High-res. Fast. Cinematic. Native lip-sync.

Follow + Retweet this post to get 800 credits sent to your DMs.

[Gregory Luneau @gluneau]

@Melt_Dem Take a look at “Ender’s Shadow”, same author, it’s from Bean’s point of view. I still think a lot about it to this day.

[Gregory Luneau @gluneau]

More proof of work.

80%-90% localization coverage reached!
https://t.co/sVtVlN9xtE