Zach Firestone

this is how you lose the room still you. just not a robot. Tono · Safer / Funnier / Custom code tonofriend → tonoit.com

Introducing LLM-PQR: provider-neutral model selection with evidence. Bring your models and priorities for cost, speed, and quality. Get an explainable recommendation; enforce local-only when needed. Public alpha: no provider calls or telemetry. GitHub.com/Amazed-Labs/llm-p…

SELL YOUR MAC MINI: Cursor is finally stepping outside the coding sphere. This product has spread like wildfire internally and I can't wait for everyone to try it. It flips how you interact with agents, replacing many temporary threads with long-term assistants. I have an executive assistant named Wonton who's been clutch on my China trip. The Forbidden Palace ticket drop sells out in minutes and I was panicking, struggling to get WeChat to cooperate. I had told Wonton through the mobile app that I was going for that day's drop, and he managed to navigate the site and secure the tickets that I wasn't able to. Oh, and he booked all my hotels! Each agent gets their own laptop to keep. We provide that in the cloud. They can talk to each other, give you control of the computer when you need it, and do much more. The team cooked here. Let us know what you think!

every lab will build a social workspace for its own agents. buzz is the workspace for everyone’s.

The Anatomy of a Hermes Agent 🤖 🧠 Brain (LLM) The thinking engine. Can be anything — Claude, GPT, Gemini, DeepSeek, local Qwen, even a mix. The model picks what you can do and how well you reason. 🦴 Skeleton (Core Loop) run_agent.py — sends messages to the LLM, gets back text or tool calls, dispatches the tool calls, appends results, repeats. Simple but relentless. 🖐 Hands (Tools) 50+ tools across 20+ toolsets: terminal, file I/O, web search, browser automation, vision, code execution, delegation, cron scheduling, messaging, Spotify, smart home... Each tool is a file in tools/, auto-discovered via registry. 🧬 DNA (Skills) Reusable procedures the agent learns. Written as markdown, loaded into context on demand. Over time the agent accumulates domain expertise — like knowing your infrastructure inside and out. 💾 Memory (Multi-tier Architecture) Fully customizable memory layers. Set up as many tiers as you need — from quick in-session notes to deep semantic long-term storage. Plug in your preferred providers, mix and match backends, build the memory architecture that fits your workflow. Your agent remembers what matters, forgets what doesn't. 🫀 Heartbeat (Cron + Scheduler) Durable background jobs that run on schedule — health checks, monitoring, recurring tasks. Survive restarts. 🌐 Nervous System (Gateway) The messaging layer. Connects to WhatsApp, Telegram, Discord, Slack, iMessage, email... Same agent, every platform, full tool access. 👥 Reproductive System (Profiles) Multiple independent agent identities on one machine. Different models, skills, memories, permissions. Each profile is a distinct "person" with its own personality and capabilities. 🦴 Reflexes (Safety) Command approval gates, secret redaction, PII protection, platform-specific tool restrictions. Won't run destructive commands without asking. 🫁 Lungs (Context Compression) When the conversation gets too long, older turns get summarized to stay under the token limit. Prevents suffocation. That's the whole organism — from skeleton to reproductive system! 🤖 Open source. Self-hosted. Yours. #HermesAgent #NousResearch #AI #OpenSource #AgentFramework

AI will take your job because you were operating as AI your whole life, but a slower version of it. Lean into your creativity and out of predictable patterns. That's one thing AI won't replace.