Hosted Hermes Agents — Your Own Hermes, Isolated in the Cloud
Hosted Hermes Agents
Run your own NousResearch Hermes agent inside Divinci — each one isolated in its own Cloudflare Sandbox container. Chat with it in the app, or connect a local Hermes, the desktop app, or any OpenAI-compatible client through a per-agent proxy URL. Bring your own provider key.
Create your Hermes agent → Talk to us
⚡ Deploy your own on Cloudflare
Prefer to self-host? Hermes Agents are built on our open-source hermesworkers Worker — deploy it to your own Cloudflare account. (Advanced: the container runtime needs Cloudflare Containers and a few secrets; see the repo README.)
One agent, one sandbox
Every hosted agent is its own tenant. There is no shared process, no shared filesystem, and no way for one agent's traffic to reach another's container.
Isolated by default
Each agent gets one Durable Object and one Cloudflare Sandbox container. The Hermes gateway runs as a non-root user; containers are lazy — provisioning is free until the first chat, and they sleep-evict when idle.
Chat in-app
Give an agent a name, a model, and a persona. It streams OpenAI-style responses right inside the Divinci web app, metered to your workspace wallet. The Divinci Agent can even set one up for you.
Connect anything
Each agent has a proxy key (hsk-…) and an OpenAI-compatible URL. Point a local Hermes gateway, the Hermes desktop app, Open WebUI, LibreChat, or your own code at it — the full /v1/* surface is proxied to your container.
New to Hermes?
Hermes is NousResearch's open, tool-using agent — the open-weight models are the brain, the Hermes Agent CLI is the body. Here's the team behind it on running open models in production.
Meet Hermes — NousResearch. Video © NousResearch.
How it works
Two trust boundaries, both enforced server-side. Your proxy key identifies an agent; it never grants access to anyone else's.
- You create an agent under your workspace. Divinci mints two credentials: a stable internal agent key (routed to the Worker) and a customer-facing proxy key (
hsk-…) you can share with clients and revoke by regenerating. - A request arrives — from the in-app chat, or from any OpenAI-compatible client hitting
/api/v1/hermes-proxywith yourhsk-key. - Divinci resolves the agent from the key, server-side. The caller never supplies an agent ID, so there is no way to pivot into another tenant's container.
- The request is forwarded to the hermesworkers Worker over a service-authenticated channel, which routes it to your agent's Durable Object and Sandbox container.
- Hermes answers — using the platform provider key, or your own bring-your-own-key when you've set one. Usage is metered to your workspace wallet, attributed to the acting user.
Connect your local Hermes in two lines
The proxy speaks the OpenAI API. Anything that talks to OpenAI talks to your agent.
export OPENAI_BASE_URL="https://api.divinci.app/api/v1/hermes-proxy"
export OPENAI_API_KEY="hsk-your-agent-proxy-key"
# ...or drop the same URL + key into a local Hermes gateway
export GATEWAY_PROXY_URL="https://api.divinci.app/api/v1/hermes-proxy"
Secure by construction
The boundaries that matter, enforced in code and covered by tests.