A big thank you to @lcnxyz for joining us on our call tonight and talking all things @almnc_ai

This weekend the team brought 15 stations online across Lyon, France. Three days of work. Around five minutes end to end for each site. Deploying computer vision to a physical location usually means weeks of integration. New cameras, a site visit, cabling, sign-off. Every one of these 15 stations went live on the cameras it already had. No new hardware, no rewiring, no site visits. We point Manako at the existing feed and it starts watching and alerting. The reason we did 15 at once was to see how deployment holds up across different setups. No two stations are the same, camera angles, how the feeds are wired, the lighting, the layout. Doing a batch like this confirms the system handles real-world variance, rather than just guessing from one clean pilot. We have our baseline, and we can now take this speed and scale the rollout to any site. Real World. Real Action.

Early demos of Leadpoet’s outreach functionality have been fun. You can almost see the moment it clicks: “Wait, this can find the right accounts, understand why now, find the buyer, and run the outreach?” Yes. That’s the point. Automate the painful parts of sales so your team can focus on the part that actually matters: Closing.

"If you start with crypto, then it's over." That is how @MaxSebti explains @webuildscore to investors outside Bittensor. First a company they can work with, and behind it a lab of 256 teams competing to build the best model. Fifteen minutes from his Beyond Finance call, on Score Studio, @manakoai, and why growth comes before buybacks.

The $200 Claude plan is a deal for one person. A company pays full price for every token. @peytonspencer on what the Ditto app does about it: retrieve the right memory and build the context packet, so a smaller model has what it needs to finish the job. From the @SubnetSummerTAO AMA, thanks for having us on.

Huge thank you @MaxSebti for joining us on our call this evening. 🙏

His key ran out mid-demo. One setting. Back to work. @peytonspencer's own GLM key hit insufficient credits halfway through a bug fix. Not planned, and he left it in. Configuration, billing, set it to both, save. The endpoint now uses his key while it has credit and Ditto billing when it doesn't, and the same prompt runs again without restarting the session. Bring your own keys, use Ditto billing, or both, per endpoint. The agent doesn't stop because a key did.

Ditto wrote me a gmail

@heydittoai v2 coming together nicely...👀

Every file you give Ditto is stored on Bittensor. So is the subnet’s growing synthetic dataset that will train Ditto LLMs. Two different things, both on Hippius. Your files sit on @Hippius_cloud. Have your own Hippius account? Connect your bucket and they go there instead, under your own keys. The subnet’s agent traces sit on Hippius too. That data is synthetic, produced by the subnet, and it is what future Ditto models get trained on, built for memory at low cost and high accuracy. 186 GB and growing fast. None of it is your data. We don’t train on user data. When Ditto wins, Bittensor wins

"Catch me up on the Subnet SummerTelegram chat from the last couple of days." 322.4K tokens read. $0.01. Three days of @SubnetSummer in six paragraphs, with the people named and the argument kept. Telegram is live in Ditto. Connect the group and the history becomes memory.

A good question about any subnet: is there a business under the mechanism, or only a mechanism? Ditto's article answers that in its own words. A product people pay for. Miners who are paid to make it better, and whose work ends up in it. And a company that pays other subnets, in production, for what it uses.

Most AI tools make you repeat yourself. Ditto remembers.

Ditto shipped their weekly update. 50 new signups this week, 2,560 registered users total, 18.8% weekly active user growth. 50 signups a week doesn't sound massive, but it's recurring. Each new user adds demand to storage subnets for file handling, search subnets for querying chat history, and inference for running agents. 50 a week is 200 a month. 200 new users pulling resources from multiple subnets every month. Sustained network demand, not a spike. Most projects talk about integration. @heydittoai is showing it in real time, one Monday at a time.

I've been pounding the @say_gm_ servers with many agents and it's holding up 👏🏼

Riley runs a small coffee roaster out of a garage. By the end of Riley's first two minutes in Ditto, it has written the prep list for the busiest week of the year. One paragraph to start: who built the store, who does the labels, and a big wholesale order that locks on October 15th. Ditto picks out what's worth remembering as you talk. Then: what's happening in October? Ditto lists all of it, asks how early you want reminding about the 15th, and when the fall market on the 17th goes in, it spots that the 15th to the 17th is the busy stretch and offers a prep checklist. One tap, and it's there, with dates. That's from one paragraph. It starts from your first conversation. Beta's open All of that from one paragraph, in the first conversation. Beta's open, link below 👇

The Bittensor $TAO ecosystem weekly update Vol.26 Half a year of Bittensor weekly updates. A 🧵

Powered by Bittensor.

Who Will Own the Intelligence of Tomorrow? One question could become one of the most important of our time: Will artificial intelligence become a resource controlled by a handful of companies, or an open infrastructure that thousands of players can build on? In 2026, a real strategic battle is becoming increasingly visible. For a long time, the argument seemed obvious: open models would remain several generations behind the best proprietary labs. That argument is becoming much harder to defend. Open-weight models are rapidly closing the gap with the frontier, costs are falling, enterprise adoption is growing, and China is accelerating massively in this direction. And now, one particularly important player is also defending this ecosystem: NVIDIA. @JensenHuang’s position is fascinating, especially because the economic incentives are almost opposite: OpenAI / Anthropic → monetize access to intelligence. NVIDIA → make intelligence abundant so that increasingly more compute is needed to produce and use it. Huang has also raised a fundamental issue: relying on an extremely small number of models creates single points of failure. Behind all the noise surrounding AI, we may therefore be witnessing two visions of the future beginning to collide. Centralized AI → a handful of labs → a handful of models → a handful of APIs → control over access to intelligence. versus Open and decentralized AI → open-weight models → independent researchers → distributed compute → permanent competition → open markets for intelligence. And this is precisely where Bittensor becomes particularly interesting. Bittensor takes this idea much further than simply publishing model weights. The goal is to build an open market where different forms of intelligence, models, data, and computational resources can compete and be rewarded according to the value they provide. There is also an interesting parallel with Bitcoin. Bitcoin demonstrated that an open protocol could coordinate an enormous amount of computing power distributed across the world, without a central company owning all those machines. Bittensor explores another possibility: What if that same logic of global economic coordination could be applied to intelligence? GPUs, data centers, developers, researchers, and entrepreneurs distributed around the world can participate in the same network and put their resources to work across specialized markets. Bitcoin coordinates compute to secure a monetary network. Bittensor seeks to coordinate global resources to produce useful intelligence. The battle ahead could therefore go far beyond the question of which lab currently has the best model. The real question may be much bigger: Who will own the infrastructure of tomorrow’s intelligence? A handful of companies? Or an open global network that anyone can contribute to? We may be witnessing the first cracks in the ultra-centralized model of AI. ➡️ bittensor.com $TAO is Not Just a Crypto Token ➡️x.com/CryptoZPunisher/status… Bittensor: A Simple Explanation of the World’s First Decentralized AI Supercomputer ➡️x.com/CryptoZPunisher/status…