ILIA.ai

If u are 0human Corp, i recommend to try this

You conduct. AI Plays. @iSales_ai

🏁 AI Video Production by me 🏁 Every business, company, and personal brand needs video content… or so I thought today, and decided to build myself a team of agents for video production. 🏆 Agent Playbook in one repo: A-1: input the product → analysis → tasks → video ideas with potential scoring A-2: selected idea → video concept, script, editing techniques, hooks A-3: prompt generation for assets: Character Sheet, Environment, Reference, Moodboard A-4: final prompt refactoring for the video generation model Manually or via MCP: ——→ Generate assets in GPT-2 ——→ Drop the assets + prompt into Seedance and get a 15-second video [Get blown away by the result, repeat endlessly, or schedule it to run automatically] Total work time: 20 minutes Dopamine level: x10 Follow for more. Results in the next post!

Want an autonomous AI-agent factory that writes code 24/7 in the cloud without you being involved? Here’s a professional, highly optimized stack that actually works — and keeps all your Claude/Codex subscriptions evenly loaded with tasks from your roadmap or autonomous routines. Layer 1: VPS (DigitalOcean, 8 vCPU / 32 GB) The muscle behind the whole operation. Ubuntu 24.04, Docker, Postgres, git worktrees — every agent gets its own branch, its own dev server, and its own Chromium for Playwright. Don’t use a Basic droplet: on shared CPU, Next.js builds will choke each other out. Layer 2: dotfiles + GNU Stow A single bootstrap.sh from a GitHub repo turns a fresh Ubuntu box into a battle-ready workstation in 15 minutes. Stow lays out your configs as symlinks — no magic, everything is visible through ls -la. Delete the droplet → spin up a new one → run the script → you’re home. Layer 3: Claude Code Native installer, not npm. Run it as a regular user, never as root. Stop / Notification hooks ping a Telegram bot — you get a push on your phone when an agent finishes or waits for your “yes.” Layer 4: tmux Persistent sessions that survive any disconnect. Naming convention: claude-- — tmux ls instantly reads like a roster of your agents. Layer 5: amux A web dashboard on top of tmux. It parses ANSI output, sees the status of every agent (working / waiting / stuck), and a self-healing watchdog catches context overflow and restarts things automatically. Session cards, peek mode, kanban — all in the browser. Layer 6: Tailscale A mesh VPN over everything. SSH/mosh go through a private network, the amux dashboard is not exposed to the public internet, and your phone uses the same URL as your Mac. Open the PWA on the go, see that the frontend agent finished a ticket, and approve the merge with one tap. A typical daily workflow — or you can set it up for the whole week: ☕ Morning — add 5 tasks to the amux kanban from your Mac 🚇 During the day — Telegram says “migration approval needed” → approve from your phone 🍝 Lunch — peek in the browser and check progress 🌙 Evening — merge completed branches, while the rest keep running overnight On top of this stack, you can build autonomous factories for mobile/web development, content production, and cloud assistants for operational tasks. I’m convinced this is the future gold standard for zero-human companies.

github.com/mixpeek/amux @mixpeek

I want to share with you the best Open-source control plane for AI agents. It's 160 ⭐️ on github today, but it's a pure gem!

@banteg thanks for new targets, bro

remaining layerzero 1 of 1 bridges theoriq (base), orderly network (arbitrum, base, bsc, orderly, solana), zentry (bsc), swell network (swell, zircuit), almanak (base, bsc), anyone protocol (base, peaq), lightlink (lightlink), over the reality (peaq), river (base, bob, bsc, sonic), lily's coin (solana), nifty island (solana), aethir (beam, solana, zksync), aegis (bsc), aori (megaeth, monad), chill house (solana), cryptocurrency coin (solana), cybro (blast), gonnamakeit (arbitrum, base), initia bridge (arbitrum, initia), kelp dao (arbitrum, avalanche, base, ink, mantle, scroll, swell, unichain, zircuit, zksync), metastreet (blast), relend network (swell), unstable coin (base), vana (base), wspn (tomo), xavier renegade angel (solana), 47 eagle (base)

THIS GUY JUST DROPPED A 16 MIN TUTORIAL ON USING GEMINI 3.1 + SEEDANCE 2.0 TO BUILD CINEMATIC $10K WEBSITES

Introducing OpenMythos An open-source, first-principles theoretical reconstruction of Claude Mythos, implemented in PyTorch. The architecture instantiates a looped transformer with a Mixture-of-Experts (MoE) routing mechanism, enabling iterative depth via weight sharing and conditional computation across experts. My implementation explores the hypothesis that recursive application of a fixed parameterized block, coupled with sparse expert activation, can yield improved efficiency–performance tradeoffs and emergent multi-step reasoning. Learn more ⬇️🧵

@FlyaKiet @superset_sh congrats team! awesome! does it kanban-style task? can agent manage them?

holy @superset_sh v2 sidebar

new claude design is amazing. just built my dream website!

@claudeai @grok how to make it works?

We're bringing the advisor strategy to the Claude Platform. Pair Opus as an advisor with Sonnet or Haiku as an executor, and get near Opus-level intelligence in your agents at a fraction of the cost.

@haydenzadams @MetaMask We’ve already started using it in Plasma and our AI DeFi Agents, much appreciated. It would also be great if you could provide an API for v3 and v4 pool data.

The Uniswap API was added to @MetaMask's meta-aggregator a month ago 🔥 Its now winning ~40% of their swaps And we're giving out API keys to all developers for free

@zmanian Are there any current examples in capital management where agents are profitable and perform better than a coin flip?

My take is vaults get entirely replaced by wallet that enable controlled to Ai agents within 6 months. There is a zero reason to aggregate capital under a manager now. You can provide risk and modeling endpoints that agents buy intelligence from mpp and x402. Eliminating vaults massively reduces the attack surface exposed by Drift and Resolv.

@Gregorein he just forget to add: "make no mistakes" in his prompt

so... I audited Garry's website after he bragged about 37K LOC/day and a 72-day shipping streak. here's what 78,400 lines of AI slop code actually looks like in production. a single homepage load of garryslist.org downloads 6.42 MB across 169 requests. for a newsletter-blog-thingy. 1/9🧵

do this to protect yourself against supply chain attacks $ cat ~/.npmrc min-release-age=7 $ cat ~/.config/uv/uv.toml exclude-newer = "7 days"

holy shitt, somebody at OpenAI leaked the entire codex codebase.. github.com/openai/codex