From sign-up to a running, supervised agent. You need a payment method and the code that starts your agent.
1
Create a workspace. Sign up, verify your email and add a card. You are not charged until a server runs.
2
Choose a template. In the console, open Agent servers → Deploy and pick Python, Node.js or Docker. Templates come with a supervised process, secrets and logs wired up.
3
Pick a region and size. Choose the region closest to your users or data. A 1 vCPU / 2 GB Basic server suits most chat and support agents.
4
Set the entrypoint. The command that starts your agent, for example python agent.py.
5
Add secrets. Add MODEL_API_KEY and any tool tokens. They are encrypted and injected when the process starts.
6
Deploy. Provisioning takes a short while. When the status shows Running, open the Agent tab to see live logs.
Prefer the API? The same flow is one request. See API (preview) in the sidebar.
Get started
Templates and entrypoints
A template is a server image with the Byte Roast runtime preinstalled. Your code goes on top.
Template
Default entrypoint
Good for
Python agent
python agent.py
Most agent frameworks
Node.js agent
node agent.js
Slack and webhook bots
Docker
Your image’s CMD
Anything else
The entrypoint should run in the foreground and keep running. If it exits, the restart policy decides what happens next.
FROM python:3.12-slim
WORKDIR /app
COPY . .
RUN pip install -r requirements.txt
CMD ["python", "agent.py"]
Configure
Secrets
Keep model keys and tool tokens out of your code. Secrets are encrypted at rest and exposed to your process as environment variables.
1
Open the server, then the Agent tab.
2
Under Secrets, add a name in capitals, such as MODEL_API_KEY, and its value.
3
Restart the agent to pick up the new value. Your image is never rebuilt.
import os
api_key = os.environ["MODEL_API_KEY"] # injected at start
Values are shown once when you save them and never displayed again. To change one, overwrite it.
Configure
Restart policies and logs
The runtime supervises your process. When it exits, the restart policy decides whether it comes back.
Policy
What happens on exit
always
Restart after any exit, including a clean one
on-failure
Restart only after a non-zero exit code
never
Stay stopped until you start it
Restarts back off after repeated failures so a crash loop cannot thrash the server. Every exit and restart is recorded in the activity log with its exit code.
Logs. Anything your agent writes to stdout or stderr appears in the Agent tab with a timestamp and level. Use your language’s normal logging; no SDK is needed.
Configure
Networking and firewalls
Servers in the same region share a private network. Firewalls decide what reaches each agent from outside.
1
Create a firewall under Networking → Firewalls.
2
Add inbound rules for the ports your agent serves, for example TCP 443 from anywhere for a webhook.
3
Leave SSH (TCP 22) open only to your own IP addresses.
4
Apply the firewall to one or more servers.
Agents that only call out to model APIs need no inbound rules at all.
Operate
Backups and snapshots
Two ways to keep a copy of a server, for different jobs.
Backups
Snapshots
Taken
Automatically, weekly
When you choose
Kept
The most recent copies, rolling
Until you delete them
Price
20% of the server price
$0.07 per GB per month
Use for
Recovering from mistakes
Before risky changes, or to clone
Operate
Billing
Servers are metered hourly and capped at their monthly price. Invoices are issued monthly.
A powered-off server is still billed because its resources stay reserved. Destroy it to stop charges.
Usage to date and the month-end projection in the console are estimates. The invoice is final.
Calls to model providers are billed by them, not by Byte Roast.
If a payment fails, you have time to update your card before servers are paused.
Reference
API (preview)
Everything in the console is available over HTTPS with an API key. Create a key under Settings → API keys. The API is in preview and may change before launch.
