Pick your agent. We host it.
OpenClaw, Hermes — and your own n8n instance. Always-on dedicated workspaces, LLM credits included. From $19.99/month per agent, cancel anytime.
~30s
workspace boot350+
LLM models24/7
designed for 24/71:1
isolated workspacesEverything an agent needs to run unattended
AgentRoost is opinionated about the stuff that hurts when you self-host. We solved it once so you don't have to.
Persistent memory you can trust
Your agent's state — vector stores, knowledge bases, conversation history, scratch files — lives on a real NVMe SSD attached to the workspace. Restarts keep everything. No "stateless" caveats.
A public URL per workspace
Every workspace gets its own subdomain under agentroost.app. Webhooks, dashboards, anything you serve from inside the workspace is reachable on HTTPS without you setting up a reverse proxy.
Predictable, dedicated resources
Intel Core Ultra cores, DDR5 ECC memory, and NVMe storage allocated to your workspace — not the usual shared-VM lottery. Performance you can plan around.
A safety net for the outbound side
A three-tier egress firewall (Spamhaus refresh daily, RFC1918 blocks, cloud-metadata blocks) keeps your agent from accidentally hitting things it shouldn't — internal networks, infrastructure metadata, abusive endpoints.
Two agent runtimes, preinstalled
Every workspace ships with both Hermes and OpenClaw ready — memory, scheduling, and connector hooks already wired up. Pick your framework in the launch wizard, bring your prompts and keys, and start.
Simple monthly billing
One price per agent, billed monthly until you cancel. No per-second surprises, no committed annual contracts, no surcharge per region. Cancel and refund pro-rata via the Refund Policy.
What people are running on AgentRoost
The agents that actually justify a persistent runtime — not the ones that fit in a Lambda.
Always-on inbox triage
Hermes watches a shared mailbox, drafts replies, routes the rest, and remembers each thread context-aware across days. Outperforms tools-only agents because it keeps state.
Code-watching assistants
An agent subscribed to a Git repo: opens PRs, replies to review comments, runs migration drills, keeps an internal changelog. Persistent memory keeps it consistent with past decisions.
Customer research and data sync
Long-running researchers that grow a vector store across sessions, ingest scheduled feeds, and answer with citations. The 1-hour-cold-start anti-pattern is gone.
Scheduled background workers
Daily summaries, weekly digests, periodic webhook reconciliation. Cron in your workspace, no separate scheduler service.
Customer-facing chatbots with memory
Real conversational agents on your subdomain, with persistent per-user threads, not the usual stateless chatbot widget.
Anything an agent can do
Hermes and OpenClaw are both thin runtimes around the LLM call loop. Whichever framework you pick, if it has a config block, an agent can run on AgentRoost.
The specs, in plain numbers
What you actually pay for. Hardware refreshed regularly so a workspace today gets current-generation silicon.
Three steps to live
No infrastructure homework.
Pick a plan
Plans from $19.99/month per agent, LLM credits included. You can cancel any time from your account.
We provision
Container starts on dedicated hardware. Persistent disk is attached. Your public URL is ready in seconds.
Drop in your agent
Browser terminal. Configure your code or use a bundled runtime — Hermes or OpenClaw. Start the loop. Walk away.
Privacy and security by default
No special enterprise tier required. Every workspace ships with the defaults that respect your data.
Dedicated workspaces, no shared tenancy
Each agent gets its own isolated workspace — its own disk, its own CPU slice, its own network namespace. No noisy neighbours, no cross-tenant data leakage, no surprise contention from the workload next door.
Your data, your rights
You can export everything we hold about your account as JSON, change your email, or permanently delete the account (and the workspaces and disks attached to it) from inside settings. Each action is gated by an email confirmation.
Encryption end-to-end
TLS for everything over the wire. Disks encrypted at rest. Workspace secrets stored via the standard tenant-key-encryption pattern, not in plain text.
No model training on your data
AgentRoost is the runtime. We don't use the contents of your workspaces — prompts, memory, files, conversations — to train any model, ours or anyone else's.
Common questions, short answers
How is this different from a serverless function or a Lambda?
Serverless wakes cold, has a hard time-out, and has no real disk. AgentRoost workspaces stay running and keep state on a real NVMe disk. If your agent benefits from "remembering" anything between calls, that's the difference.
How is billing handled?
Simple monthly subscription, from $19.99 per agent with LLM credits included. The card is charged on the same day each month until you cancel. No per-second meter, no surprise overage, no annual lock-in.
Do I share resources with other customers?
No. Each workspace gets its own isolated slice of CPU, memory, disk, and network. Your agent's performance is not affected by what the workload next door is doing, and there is no shared filesystem.
Which agent frameworks do you support?
Three today, picked in the launch wizard. Hermes (our proven default) and OpenClaw (the popular open-source agent, hardened for hosting) are fully managed Telegram agents. n8n runs as your own workflow-automation instance — your login, your data — provisioned on a dedicated workspace. All three get the same always-on workspace and included LLM credits.
Can I bring my own LLM keys?
Yes. Hermes and OpenClaw read your model credentials from the workspace environment. n8n is your own instance — add any provider credential straight in its editor, or use the pre-wired one backed by your included credits. AgentRoost doesn't intermediate between you and the model provider.
What happens to my data if I cancel?
You can export it before cancelling. After cancellation, the workspaces and their disks are torn down per the Refund Policy. We don't silently archive anything.
Do you support multiple agents on the same workspace?
A workspace is provisioned for one agent. If you need a second agent, it's a second workspace at the same per-agent price. This keeps blast radius small if one agent misbehaves.
Ready when you are
One subscription, one workspace, all the moving parts already wired up.