Pethost for Python

The Python SDK for Pethost, hosting for Docker Compose projects on a machine of your own. It has the 14 calls of Pethost's API, the ones an agent has through its MCP server: read the machine and its projects, deploy, read logs and HTTP requests, run commands, move files. It has the API's streams too, which follow an operation, the logs and the requests as they happen. Everything is typed, there is a synchronous client and an asyncio one, and the whole package is generated from the API's definition, so its methods, its fields and their documentation are the API's own.

Install

pip install pethost

Python 3.10 or newer. With uv: uv add pethost.

A token

The SDK signs in with an API token. Make one in the panel's Settings, under "API tokens", and give it to the SDK in the environment variable PETHOST_TOKEN, or as the client's first argument. A token acts as you: keep it out of your code and your repository.

A first program

It reads the machine and its projects, then deploys to the project notes by writing one file, /hello.txt.

from pethost import FileChange, Pethost

with Pethost() as pethost:  # The token: PETHOST_TOKEN.
    answer = pethost.get_machine()
    machine = answer.machine
    if machine is None:
        raise SystemExit("the answer has no machine")

    print(machine.hostname)
    for project in answer.projects:
        print(project.project_id)

    deployed = pethost.deploy_project(
        "notes",
        files=[FileChange(path="/hello.txt", text="Hello from the Pethost SDK\n")],
    )
    for violation in deployed.violations:  # The file was refused: nothing was deployed.
        print(violation.location, violation.violation_message)

    if deployed.operation is not None:
        print(deployed.operation.status.name)
    if deployed.project is not None:  # The project as it is once the deploy ended.
        print(deployed.project.url)

AsyncPethost has the same methods for asyncio:

import asyncio

from pethost import AsyncPethost


async def main() -> None:
    async with AsyncPethost() as pethost:
        print(await pethost.get_machine())


asyncio.run(main())

Methods

A read-only call changes nothing; a destructive one may delete or overwrite something; a stream returns its responses one by one, as "Streams" below shows. A request's fields are its method's keyword arguments, a message is a frozen dataclass of the same name, and each docstring is the API's own description: help(Pethost) and your editor show it. What the API calls absent is None where a type allows None, and the zero value anywhere else.

Streams

A stream is a method that returns a generator: read it with for, and each response comes as the API sends it, for as long as the loop runs. This program prints a project's newest log lines, then follows them.

from pethost import Pethost

with Pethost() as pethost:
    # The newest lines, and where a stream goes on from them without missing one.
    newest = pethost.query_container_logs("notes")
    for line in newest.lines:
        print(line.service, line.text)

    # Each later line as it is written. The loop runs until the stream fails, which raises; to
    # stop it sooner, leave the loop.
    for response in pethost.tail_container_logs("notes", after_cursor=newest.tail_cursor):
        if response.line is not None:
            print(response.line.service, response.line.text)

The call is made when the loop starts. A stream that fails raises its PethostError from the loop, once, after the responses that came before it; one that ends by itself just ends the loop. Leaving the loop ends the call: Python closes a generator that nothing else holds. Where a variable holds it, its close() ends the call, and so does contextlib.closing around it. The client's timeout does not bound a stream, and the package does not call again by itself: where a response carries a cursor, as a log line does, a new call goes on from it.

While this client waits for a response, Python does not act on Ctrl-C: KeyboardInterrupt is raised when the API next sends something. That is the next response or, on a quiet stream, what the API sends through a silence to keep the stream open, which the loop never sees. A program that must stop at once reads the stream with AsyncPethost.

AsyncPethost has each stream as an asynchronous generator, read with async for:

import asyncio

from pethost import AsyncPethost


async def main() -> None:
    async with AsyncPethost() as pethost:
        async for response in pethost.tail_container_logs("notes"):  # From now on.
            if response.line is not None:
                print(response.line.service, response.line.text)


asyncio.run(main())

Leaving that loop ends the call on the event loop's next turn. Cancelling the task ends it too, so Ctrl-C stops the program at once, and asyncio.wait_for can bound the reading by time. Where a variable holds the generator, await its aclose(), or put contextlib.aclosing around it.

Errors

A failed call raises a PethostError, and its class is the code: CanceledError, UnknownError, InvalidArgumentError, DeadlineExceededError, NotFoundError, AlreadyExistsError, PermissionDeniedError, ResourceExhaustedError, FailedPreconditionError, AbortedError, OutOfRangeError, UnimplementedError, InternalError, UnavailableError, DataLossError, UnauthenticatedError. error.message is prose for people. What a program branches on, beside the class, is error.detail: ProjectBusy, ProjectChanged, MachineUnreachable, MachineUpdatePending, NoMachine, or None.

from pethost import NotFoundError, Pethost, PethostError

with Pethost() as pethost:
    try:
        project = pethost.get_project("notes").project
    except NotFoundError as error:
        print(error.message)  # No such project: the message lists the ones there are.
    except PethostError as error:  # Any other code.
        print(type(error).__name__, error.message, error.detail)

A wrong argument the SDK sees itself, such as two members of one oneof, is Python's own TypeError, ValueError or OverflowError, raised before anything is sent.

JSON

Every message has the API's own JSON form, the one the API speaks over HTTP and its reference documents. to_dict() gives it as a dict for json.dumps, and the class's from_dict() reads one back.

import json

from pethost import Pethost, Project

with Pethost() as pethost:
    project = pethost.get_project("notes").project
    if project is not None:
        text = json.dumps(project.to_dict())  # {"project_id": "notes", ...}
        print(Project.from_dict(json.loads(text)) == project)  # True

A key is the API's name of a field, an enum is its value's full name, a 64-bit integer a string, a time RFC 3339 and bytes base64. What is absent is left out, and so is a zero where the type has no None: the API may write such a zero, and reads its absence as the same.

from_dict refuses a key that is no field of the message with a ValueError, as the API does, so a misspelt key cannot pass for an absent one. A field that a newer API added is such a key to an older version of the package. It takes what the API takes beside its own form: a key in lowerCamelCase, a 64-bit integer as a number, an enum by its number. An enum's value that this version of the package has no name for is its number, both ways. dataclasses.asdict is not this form: it keeps a datetime and an enum's member as Python has them.

Links

• The API's reference: every call, message and field.
• The panel: your machine, your projects and your API tokens.
• pethost.dev: what Pethost is and what it costs.

Docker Compose hosting at one flat price

Pethost is managed hosting for Docker Compose projects. A subscription rents one virtual machine that only your account uses. Pethost's software on it deploys your projects, gives them HTTPS addresses, keeps their request logs and backs them up every night. You run it through your agent or from a web panel: both show the same projects, logs and requests.

• A machine of your own: nobody else's projects on it. No cold starts, nothing sleeps. Security updates install themselves.
• HTTPS addresses: anything.yourname.pethost.app at once, or your own domain with one DNS record. Certificates renew themselves. A password page in front of the whole site: one sentence to your agent.
• Deploys from GitHub: every push deploys itself, and an earlier commit is one step back.
• Request logs: every request with its status and time, the busiest paths and the error rate, kept for 30 days.
• Nightly backups: encrypted, stored off the machine, kept for 14 days.
• SSH and SFTP into any container, and a terminal in the panel.
• Docker Compose, as written: a project is a Compose file, and the same file runs anywhere else Docker does. A database is one more service in it, at no extra cost.
• In the EU: your data lives on your machine, and requests to your sites go straight to it, never through our panel.

Pricing. Starter is €9.99 a month for 2 vCPU, 3 GB of memory and 30 GB of SSD. Grow is €17.99 for 4 vCPU, 6 GB and 60 GB. Scale is €59 for 8 vCPU, 30 GB and 200 GB. Prices are without VAT. As many projects, containers and databases as fit: nothing is metered, and nothing is billed on top. There is no free tier. See the plans.

Who it is for. Solo developers and early startups who host several projects, bots or small apps and want one fixed monthly bill, and people who build with an AI agent and want it to deploy and run what it wrote. It is not for large teams whose applications need several regions or autoscaling: the machine is the limit.

How Pethost compares

About Pethost has each comparison in full, with prices.

Questions

Which AI agents does Pethost work with? Almost any agent that supports skills and MCP. The ones we tested ourselves: Claude Code, Codex, Cursor, ChatGPT, Gemini CLI, OpenClaw, Hermes and Pi.

What can I deploy? Anything that runs in Docker: web apps, APIs, bots, workers, databases. No Dockerfile? With the plugin installed, your agent knows the details and writes it for you.

Do I need an agent? No. The web panel does the same, and shows the same projects, logs and requests.

Can I use my own domain? Yes. One DNS record, and the certificate comes by itself.

Is there a free tier? No: €9.99 rents a real machine, so nothing sleeps. Cancel any time: refunds and cancellation has the rules.