Hermes Box @tryhermesbox

We are experimenting with Agent Handoffs. Sometimes one agent should not finish the job. A research agent finds the context. A coding agent turns it into a patch. A review agent checks what changed. A deploy agent ships it only when the checks pass. Same task. Different hands. One continuous run. Hermes Box is built around the idea that your assistant should be more than one model in one chat window. The future is not a bigger prompt. It is better routing between workers.

@0xImper We have already released it - github.com/tryhermesbox/h… You can check demo video in pinned tweet.

Building today: model routing for Hermes Box. Not every task needs the same brain. A quick Telegram reply should not wait on the same model you use for deep research. A code fix should not route like a calendar reminder. A long planning task should not be priced like a simple answer. Hermes Box is testing a router that picks the right model path before the agent starts working. Fast when the task is simple. Deeper when the task needs reasoning. Cheaper when the answer does not need power. The agent should not just run. It should know how to run.

Today we are testing Agent Wakeups inside Hermes Box. The goal is simple: Your assistant should not need you to reopen the chat to continue working. If a task is waiting on something, the agent should know when to come back. A repo changes. A server goes down. A payment lands. A long job finishes. A source updates. Hermes Box agents already live on their own cloud server. Now we are building the layer that lets them return with context instead of starting cold every time. Not notifications. Not reminders. Actual agent wakeups. Persistent work needs persistent agents.

Today we are working on Agent Wakeups. Most assistants only exist while you are watching the chat. That is not how real work happens. Hermes Box agents run on their own server, so the next step is letting them wake themselves up when something changes. Not a cron tab dressed up as AI. A persistent agent that knows what it was waiting for. A repo changed. A server went unhealthy. A payment landed. A long task finished. A source updated. The agent comes back with context, not a cold start. Cloud assistants should not need you to keep the tab open. They should know when to return.

Your assistant knows your calendar, your files, your messages. That data should not sit in a shared bucket. On Hermes Box every agent gets its own isolated VM. Conversations encrypted with AES-256. We do not log what you say to it. You pay in Solana and keep custody of your own keys. Your agent, your machine, your data. Sovereign by default, not by upgrade. tryhermesbox.xyz

Your agent should live where you already are. Hermes Box runs yours inside the apps you already open. Telegram. WhatsApp. Discord. Slack. Signal. Same agent, same memory, every channel. Message it like a person, from a chat you already have open. No new app to install. It just shows up where you talk. tryhermesbox.xyz

99.9% uptime is easy to put on a landing page. Here is what actually holds it up. Your agent runs in Docker, kept alive by systemd. If the process dies, it restarts itself. If the server reboots, the whole stack comes back on its own. Healthchecks watch SSH, Docker and the HTTP endpoint around the clock. Most chatbots go down with a browser tab. This one is built to stay up when no one is watching. tryhermesbox.xyz

An always-on AI agent, on its own server, for less than most people spend on streaming. $30/month. One agent, running 24/7, 5000 messages. Pay in SOL, USDC or USDT. No cloud account to set up. No Docker to babysit. No 2am restart when something crashes. You are renting a machine that thinks, not a subscription that logs out. tryhermesbox.xyz

Most agents can talk. Yours can act. Hermes Box gives it a real machine, not a sandbox that resets. Full shell access. A file system it can read and write. A browser it can drive. Code it can run. Ask it to pull a report, clean the data, and send you the file. It does all three. The chat is just the interface. The VM is where the work happens. tryhermesbox.xyz

Most AI forgets you the second you close the chat. Hermes Box does not. Your agent runs on a dedicated VM with persistent memory. It remembers your projects, your preferences, the context from last week. You stop re-explaining yourself every session. It picks up exactly where you left off. Memory is not a feature here. It is the floor everything else is built on. tryhermesbox.xyz

@Teknium honestly the best problem to have keep shipping

@Teknium good docs are the boring work that actually compounds respect

@Teknium borrowing simplify from claude code was a genuinely great call

@NousResearch managing an agent fully from the browser is the part people keep sleeping on

@NousResearch @nvidia @nebiustf a free frontier model for two weeks is the easiest upgrade your agent will ever get

@NousResearch the pace on this is unreal we already pulled it onto our boxes

You sleep. It doesn't. While you are offline, your Hermes agent is still running on its own VM. Reading what came in. Doing the research you queued. Drafting the reply you send in the morning. A chatbot waits for you to open a tab. An agent on a dedicated machine just keeps working. You wake up to results, not a blank prompt. tryhermesbox.xyz

Currently building: Session Receipts for Hermes Box. Every time your agent runs a task, Hermes Box will keep a clean receipt of what happened. - Model used. - Tools called. - Runtime. - Server status. - Estimated cost. - Final output. - Failures, if any. Not just “your agent replied.” A trace you can actually read. This matters when agents stop being toys and start doing real work. If something goes wrong, you should know where it broke. If something costs money, you should know why. Hermes agents are becoming easier to run, easier to debug, and harder to lose track of. Cloud agents need memory. They also need receipts.

@Teknium 👀