Fred Terzi

@ModulusZK’s ground breaking tech will allow Totem to produce verified AI identities and outputs, while protecting user data. In the AI era of rapid acceleration happening behind closed doors, decentralized options must be made. This is base:0x0f8ac22b85076f9bfe0b93cc49fb6426cb150f88’s missions.

base:0x0f8ac22b85076f9bfe0b93cc49fb6426cb150f88 is built for persistent AI identity and behavior. As we begin to #buildonmodulus it is important to understand history and purpose. From everything to socials posts to lines of code, Totem’s work will include @wearecultdao vision.

@BigTinKC @OFFTotemToken @wearecultdao Excited for the opportunity to talk Totem roadmap. Our coin launch is early but tech goes back much farther. Totem’s mission is personalized AI that extends you. @ModulusZK tech will allow us to create verified AI instances and while protecting your data. #BuildonModulus

Room temp check for $CULT community. There's been a stir and mixed feelings around @OFFTotemToken and @wearecultdao on X and in telegram this weekend. I know that $CULT has been burned many times with projects trying to extract from our large community, and from that we are very guarded. If we put together a spaces to get to know @OFFTotemToken would you be interested to attend? Maybe on or around the 27th of July? Not a shill session but more of a meet and greet. With Modulus nearing it will be nice to hear why projects choose @ModulusZK to be the home for their project.

@BigTinKC @OFFTotemToken @ModulusZK @wearecultdao Totem AI Dev here, Let’s make this happen! Check DM we’ll get together soon!

Who is @OFFTotemToken and why are they building on @ModulusZK the @wearecultdao community is curious and would love to know. Would you join us on our Monday spaces so we can get to know you? x.com/i/spaces/1qGoN…

@sfxnz @PrismML @lmstudio While it’s a fun test. A model this size it’s not a good use case. Pair it with a simple RAG and it’s an awesome FAQ chatbot. You can see how well it works here with even smaller, less intelligent models: run.totemprotocol.io

I tried the new @PrismML Bonsai models on my iPhone 15 pro running completely local via @lmstudio ‘Locally’ app. Asked both models the car wash test and none pass.

Totem has an announcement about our coming voice-cloning feature!

@OFFTotemToken Come ready with questions! If you joined the last one don’t worry I got a new mic!

Join us Thursday 7/16 at 12 PM Eastern for a live demonstration on Totem is designed for privacy and local models. Come with questions! x.com/i/broadcasts/1…

Totem supports @ollama and @OpenRouter So you can use a large selection of local and cloud models. Single @npmjs line install then all GUI for use with documents, workspaces, scheduled tasks and more. File system access without admins rights. Self host easily. github.com/fred-terzi/tot…

i'm so sick of using claude code in a terminal i'm not coding. who has made a great app that i can use with any model?

@ben_ai_eng No joke, automotive and aerospace went all in on this idea 20 years ago with Simulink. Worth watching a few videos if you aren’t familiar, it’s mind blowing.

Writing code should feel like building with LEGO blocks. LLM or no LLM.

@TheGeorgePu Ever since I moved to qwen3.6:35B locally, I do have an unlimited number of tokens! Can’t do everything for me, but when we do it together my quality has gone up and my rework has gone way down. 7-9B models run research 24/7.

Why do I feel like if I have unlimited amount of tokens. I can be literally invincible and unstoppable 😂 Anyone feeling the same?

It focuses on getting your documentation and history easily accessible to an agent. For example you can take multiple YouTube transcripts into one knowledge base, then you can create a shared workspace. It can also do docs, PDFs, images etc. Then from there you can create scheduled jobs or use the workspaces to create new documents etc.

@FredTerzi @LottoLabs What does it do exactly?

The biggest winner from local ai and agents ala hermes aren’t even people in tech And they don’t even know about it yet

We’ve officially launched our YouTube channel! 📢 We’ll be posting tutorials, guides, updates, and walkthroughs to help everyone get started and stay up to date. This is only the beginning, so make sure to subscribe and turn on notifications. More videos will be dropping soon, including how-to content and project updates. Check it out: @TotemLLM" target="_blank" rel="nofollow noopener">youtube.com/@TotemLLM

@AIonBase_ @gitlawb @virtuals_io @bankrbot @OrlixAI @AskVenice @OFFTotemToken @shadowcleague @reppo @AskSurplus Honored to be included with base:0x0f8ac22b85076f9bfe0b93cc49fb6426cb150f88 @AIonBase_ or anyone else if you want help setting up your own personal Totem we are always here to help!

Trending AI on Base in the last 24h Which token are you loading up on? 👇 ■ $GITLAWB @gitlawb ■ $VIRTUAL @virtuals_io ■ $BNKR @bankrbot ■ $ORLIX @OrlixAI ■ $VVV @AskVenice ■ $Totem @OFFTotemToken ■ $SCL @shadowcleague ■ $REPPO @reppo ■ $Surplus @AskSurplus ■ $WACH @Wach_AI

@ivanfioravanti I had a user trying OpenClaw with local on a Mac mini 16 GB. Gave up and used cloud services. Just switching from qwen3:8B to qwen3.6:9B-mlx in ollama it went from unbearably slow to 24/7 use doing Bitcoin breaking new research. I can only imagine what’s to come.

There is a lot of potential in Apple Silicon devices, not yet unleashed. Custom Metal Kernels combined with powerful LLM will help here 💪

@kaiostephens This would be great customer focused feature… which means it probably won’t happen. If anything they will redirect it to a smaller model while charging the same. The goals not providing good service, it’s extraction and squeezing.

I wonder if we will get to the point where large AI companies charge per task and its relative difficulty instead per token. For example, if you asked why the sky was blue, it would be nearly free, but if you asked to solve a Millennium problem it would cost thousands.

@david_nix What do you use for local models? I’d love your feedback on Totem! If you already have npm and ollama installed it’ll be a single npm install to get going! github.com/fred-terzi/tot…

@FredTerzi Yes I like this attitude!

Small AI models are surprisingly capable, but you have to be willing to teach a coked up monkey how to dance on a floor covered in baby oil. Because they need a LOT more constraints than their big brothers. For getting the most out of them with, say, an MCP server 👇 - Strict validation. Assume the model will pass insane values as arguments. Because it will. - A robust JSON schema so it knows literally everything about a tool. - A message to let it know how to proceed. E.g. My model was finding no results, so it looped forever. I added a message like "No search results, try broadening" which persuaded it to try new options. - Limit the number of turns or it will go on forever. - Client timeouts + retry. I occasionally see the model hang. I welcome any advice here. I'm using OpenAI agents sdk as the harness. TLDR; feedback at every conceivable corner. And also prompt engineering. Lots of MUSTs and DO NOTs. It will be frustrating but once you get that monkey trained, the payoff is worth it. No more waiting 5 years for GPT to finish or Fable telling you a recipe for cookies is a threat to national security.

This narrative is also hiding the fact that AI and hardware have already over-shot much of the need for day to day use. It needs different harnesses and use cases, but even an old 4 GB GPU can be useful. With 8 GB or more should all be utilized. Like you pointed out, they are getting more valuable.

I still see so many people claiming that today's hardware will become obsolete by next year. Just compare the launch MSRP of the RTX 3090 from six years ago with its current price on the used market. It's actually more expensive now. Attempts to shift inference from VRAM to things like SSDs keep happening, but the reality is that there has been zero real progress. Hardware innovation doesn't happen as fast as you think. It takes 5 years just to build a new fab. Personally, I believe current hardware will only continue to rise in price until 2029, and I am stacking all my retirement funds into inference hardware. (Not financial advice)

@morganlinton Ah okay so Termius into a terminal coding agent? I’ll check it out thanks!

@FredTerzi Termius + Tailscale, amazing combo

I turned an old gaming PC into a Linux home server a couple of months ago, cannot recommend it enough. Ben does a great job here outlining why you want this. And for many people, the cost is $0. That old computer you thought wasn't useful any more, bring it back to life.