have not touched perplexity/claude/chatgpt for 2 weeks now and see no reason why i would.

running my fully private (almost) AI personal assistant stack that handles research, portfolio management, journalling, coding, etc.

here's the full stack:
- Hermes agent (agent harness)
- Venice (private, uncensored AI inference API)
- Honcho (behavioral analysis/long term memory)
- Obsidian (knowledge vault)
- qmd (on device search engine, both text/vector search)
- browser-harness (CDP browser automation, agent browses the web like a human)
- Tavily (search API)
- Codex (coding, powered by private coding models on Venice too)

all on a 16gb macbook pro.

there are currently only 2 touch points (AI inference & search API) that got offloaded to cloud -- AI inference due to hardware constraints, search API has no workarounds. but i'm opting for purely private/e2e model choices on Venice and queries get filtered by the agent before hitting the search API. this way, we keep data leakage to a minimum.

it's also pretty sick that locking base:0xacfe6019ed1a7dc6f7b508c02d1b04ec88cc21bf grants you daily refreshing compute credits denominated in base:0xf4d97f2da56e8c3098f3a8d538db630a2606a024 -- meaning i'm running private, anonymized SOTA inference daily for free.

of course, full privacy is definitely the end goal. i'm slowly working towards running everything completely local while offloading only high-complexity inference to Venice.

looking into stronger workhorses (eyeing upcoming m5 studio ultra release/refurbed 512gb options for SOTA models) to close that last gap for a complete e2e private intelligence stack.

firmly believe owning your data in the age of AI surveillance and big tech data centralization is as important as owning your own assets.

like we've said for years: "not your keys, not your coins". you don't custody all your money with a bank, why custody your thoughts and private information with OpenAI/Anthropic?

if you haven't already started, highly recommend you look into it.

remember, that's why we got into crypto in the first place. useful resources:
- hermes-agent.nousresearch.com/docs/
- docs.venice.ai/models/overview
- hermes-agent.nousresearch.com/docs/user-guid…
- obsidian.md
- github.com/tobi/qmd
- github.com/browser-use/br…
- docs.tavily.com/documentation/…
- docs.venice.ai/guides/integra…

agent = llm + planning + memory + tools + while loop

read this thread to determine if your favorite projects are workflows or agents ↓

warning: this newfound knowledge might cause you to midcurve, just like i did a couple of times @AnthropicAI llm = the brain of the agent

llm is the reasoning engine within the agentic entity that "plans"

i have no tier list, it's either an in or out for me

my personal approach focuses on backing actual bootstrapped (non-VC), AI builders who consistently ship and know how to navigate the current market

this is a personal choice because I know the GenAI field well. i have been using/building with LLMs every single day for the past two years now. i'd like to believe this gives me an edge over most.

when evaluating the CT timeline, i noticed most genuinely do not fully comprehend AI's capabilities. i'm literally here to hedge my fear of getting replaced in the next 5 years, and as a gen z, NVDA/TSMC/AMD ain't gonna cut it if i wanna make it — that's why i'm in the trenches.

we go big or go home.

i've also been in crypto long enough to see value in crypto-native founders like @jyu_eth. closed-minded tradAI builders would probably never tokenize a cent of their work on chain. if @AndyAyrey weren't open-minded enough, he would have already shut goat down at sub 10mc.

look, i get that it's not all just about the tech. i've been aping shitters like wif (hat tech) since last nov when most of you are still suffering bear market PTSD.

i fully understand there are certain things you simply can't control e.g. @pmarca shilling @truth_terminal , elon shilling PNUT. but there are elements you definitely CAN control — study error of process vs error of outcome (h/t @santiagoroel, learnt shit tons of my crypto investing from him)

what i'm really saying is without the tech, these "memes" or "agents" won't have staying power.

lemme give you the CT speak real quick:

meme = number go up
talking meme = number go up more
many talking memes = who talks better
better tech & execution = talk better
talk better = more attention
more attention = more number go up

"talking" can be done through text, images, videos, audio, other forms of data, etc.

compare @0xzerebro and @luna_virtuals , who talks better? who resonates more? who can rally the troops more effectively? now reverse engineer that and ask why.

people have no idea how many of these are simply 20mins prompt-engineered OpenAI api calls hooked up w a twitter/telegram account. most of these will go to zero in the next month, and i sure as hell ain't playing that game.

for all those asking me for quick trade advice, i suck at it btw. i'm an investor, not a trader. once i pull the trigger, i usually size w conviction — and wait. trading and investing have different playbooks, and i'd like to believe i'm currently building a robust one for this incredibly nascent vertical.

so far, i'm choosing to back @AndyAyrey [goat], @jyu_eth [zerebro] and @NickLiverman [chaos] based on the thesis that i've put out (a more detailed one in the works) — see x.com/terrytjw/statu…

s/o @shawmakesmagic as well, love what that team has been shipping with eliza, but ai16z just isn't for me (respek the pamp tho)

my brain is still severely damaged by luna collapse 2022 😂 and DeFi isn't exactly my forte anws (too many unknown variables imo) — would prefer playing in my lane of backing pure AI execution

onchain AI is here to stay.

i haven't felt this way since my last contrarian bet on sol at $12. many did the same, and what did they do? follow the devs, bet on execution.

or you can midcurve and trade TON/eth L2s lol wtv glhf

AI (and memes) is a secular trend, and we're currently only ONE month into this new vertical. my bet is that more s-tier valley AI devs will keep flowing in as they see what we are cooking rn. those will be the ones i would double, triple down on.

and when they eventually enter the space to bootstrap their agentic projects, i'm betting they won't use eliza or wtv us crypto peeps have been cookin up. they'll cook the core tech themselves — and i'm betting my ass it'll be waaay better.

i say this because i've seen how cracked those tradAI devs are. those are the people who literally put together your ChatGPTs. even as a bozo AI dev, that's i path i'll take too. we simply want maximum control over our own agentic systems. of course, it's a different story if SDKs like eliza are well-maintained and proven over the next year.

you're given the opportunity front run VCs and back actual builders in this meta right now

the choice is yours, who will you back?

terry satisfactory answer? @onlytraderock

AI-powered dev is reshaping how we build software.

Forget 10x devs. We're now talking 1000x potential. The secret weapon? AI dev tools, with one standout: @cursor_ai

Look no further because everything you need is here.

A comprehensive playbook on how to kill it with Cursor: I will continue to post updates in this thread as I learn more. Will be closely following Cursor's development over the next few months.

Bookmark the first tweet so you don't miss anything.

Sections are signposted with bullet points (➤ and •) for easier navigation.

Let's go.