Comparison · checked October 22, 2026

Ephemeral vs your own servers

Running agents yourself means an agent framework on servers or containers you operate, with the isolation, credentials, limits and records you build around it. Ephemeral runs a complete background agent task instead: your prompt, files and allowed connections go in, a fresh microVM does the work with a budget reserved up front, and the machine is destroyed when it finishes.

As of October 22, 2026, Ephemeral charges $0.005 per run plus $0.0002 per second of run time, with model calls at the provider's price plus 15%, from prepaid credits with no subscription; on your own servers you pay for the servers and your model provider directly, and build and run the rest yourself.

Side by side

Feature	Ephemeral	Your own servers
Isolation per run	A fresh microVM for every run, destroyed when it ends. Nothing carries over between runs or organizations.	Whatever you build: a shared server, a container per task or a VM per task. Separating tasks and customers from each other is yours to design and test.
Credentials	Held by Ephemeral and added outside the machine; the agent sees only each connection's name and tools, and credential values are removed from responses.	Usually environment variables or a secret manager the agent's process can read, unless you build a broker in front of every API.
Access to your APIs	Allow-lists per run: HTTP methods per API connection, tools per MCP server. A run can narrow a saved connection but never widen it.	As wide as the credential: a key that can write lets the agent write.
Web access	Off unless the run asks for it; requests go through Ephemeral's own bot, only to public addresses, with a safety check and an optional site allowlist.	Whatever the network allows, unless you add an egress proxy and rules.
Spending control	A hard budget per run, reserved before it starts; a model call or web request that would go past it is refused.	Provider-level limits for the whole account, or a metering layer you write.
Cost accounting	The exact cost of every run, split into run fee, run time, model calls and web requests.	Server bills and provider invoices, which you attribute to tasks yourself.
Run record	Every step streamed while it runs and kept for 30 days: model calls, tool calls, connection requests, output and cost.	The logs you add, kept where and as long as you decide.
Scheduling and callbacks	Cron schedules in any time zone, and a signed callback when each run finishes.	A scheduler and a notification path you run.
Patching and upkeep	Ephemeral maintains the worker image, the engine and the machines on the hosted service; a self-hosted installation updates to the releases Ephemeral publishes.	You patch the operating system, runtimes, agent framework and dependencies, and watch the servers.
Cold start	Every run boots a new machine, which takes time and is what the run fee pays for; there is no warm machine to reuse.	A long-running process answers at once; a fresh container or VM per task adds its own start time.
Cost	$0.005 per run plus $0.0002 per second of run time, model calls at the provider's price plus 15%; or the self-hosted edition from $490 a month.	Your servers, your model provider's prices and the engineering time to build and run the rest.

Which to choose

Choose your own servers when…

You need full control of the runtime: GPUs, special hardware, long-lived processes or state kept between tasks.
Your agents must run in a network with no route to the internet or to outside services.
You already operate a container platform with isolation, secrets and metering, and a fresh machine per task would add nothing.

Choose Ephemeral when…

You would otherwise build per-run isolation, a credentials broker, budgets, schedules, callbacks and a run record yourself.
Agents handle customer data or call production APIs, and you want each run sealed, allow-listed and recorded.
You want to start today with an API call, the CLI or the MCP server, and pay per run.
You need it inside your own AWS account: the self-hosted edition runs the same engine there.

What this comparison does not cover

This page describes what each approach involves, not a specific product. It does not price your servers or your engineering time, and it does not measure start-up times or throughput, which depend on your setup. If something here is wrong or out of date, tell us through the support page and we will correct it.

Sources

Ephemeral's side comes from its pricing page, whose prices it shows as they are today, and from its documentation on security and data, connections, web access, budgets and billing, runs, schedules and self-hosting.