arkitekt
Turn your Python functions into apps you can orchestrate, share, and scale.
What is Arkitekt?
Arkitekt is an open platform for building, connecting, and orchestrating
computational apps. arkitekt is its Python client: it takes your ordinary Python functions and
exposes them as remotely callable, orchestratable building blocks — without you having to write
servers, APIs, message queues, or UIs.
Declare an app, run it, and its actions become available on an Arkitekt server where they can be:
- Called from anywhere — other apps, notebooks, scripts, or the web UI.
- Composed into real-time workflows that wire your functions together.
- Given a GUI automatically, generated from your Python type hints.
- Shared with your team behind central authentication and permissions.
- Packaged and deployed as a Docker container with a single command.
Arkitekt grew out of the needs of data-intensive science (it has first-class clients for microscopy, electrophysiology and graph data), but the core is domain-agnostic — any Python workload fits.
📚 The best place to understand the platform and its concepts is the documentation at arkitekt.live.
Installation
pip install "arkitekt[all]"
This installs the arkitekt command line interface, the runtime and every service client. Prefer a
lean install? The CLI and packaging tooling are always included — pick only the extras you need:
| Extra | Brings in |
|---|---|
rekuest |
the distributed runtime run(app) needs to offer actions |
mikro |
microscopy and imaging data |
elektro |
electrophysiology data and simulations |
kraph |
knowledge graphs and measurements |
fluss |
workflows, and the engine that runs them |
kabinet |
managing deployments of apps |
unlok |
users, clients, hubs and redeem tokens (lok) |
alpaka |
LLMs and chat |
lovekit |
WebRTC streams and rooms |
serve |
serving an app from a FastAPI application (arkitekt-fastapi) |
qt |
Qt integration (arkitekt.qt) |
tqdm |
a tqdm that reports progress to the running task |
pip install "arkitekt[rekuest,mikro]"
Declaring, inspecting and packaging an app, and calling services with easy, need no runtime; offering
actions with run(app) needs the rekuest extra. arkitekt requires Python 3.11+.
Offering actions
An App is a declaration: its identifier, its version, and the actions it offers. Any function
decorated with @app.action becomes a callable building block on the platform. Its arguments and
return value are inferred from the type hints, which also drive validation, documentation and the
generated GUI. The first line of the docstring becomes the action's title, the rest its description.
from typing import Annotated
from arkitekt import App, Description, run
app = App("hello", "0.1.0")
@app.action
def greet(
name: Annotated[str, Description("Who to greet")] = "world",
times: Annotated[int, Description("How often to say it")] = 1,
) -> str:
"""Greet
Says hello.
"""
return " ".join([f"Hello {name}!"] * times)
if __name__ == "__main__":
run(app)
Nothing connects when the App is declared. run(app) authenticates (opening your browser the first
time), registers the actions and blocks until you stop it. The server is taken from $FAKTS_URL, or
passed explicitly with run(app, url="localhost"); headless=True prints a device code instead of
opening a browser.
Using a service inside an action
Actions that need a service client name it in services= and take it by annotation — arkitekt injects
the client, just like the running Task:
from arkitekt import App, Task, run
from mikro import Mikro, mikro_service
from mikro.arkitekt.specs import Volume
app = App("inspect-volume", "0.1.0", services=[mikro_service])
@app.action
def describe(volume: Volume, mikro: Mikro, task: Task) -> str:
"""Describe Volume"""
task.progress(50, "Reading")
return f"{volume.data.shape}"
if __name__ == "__main__":
run(app)
Beyond actions, an app can declare @app.model result types, @app.state that the platform publishes
live, @app.startup/@app.shutdown hooks and @app.background tasks, and a typed app context
(App(..., app_context=Setup) together with run(app, context=Setup(...))). The
examples show one each.
Calling services
Scripts and notebooks that only call the platform don't declare actions. easy declares an app for
you, connects it, and hands back the clients of the services you name:
from arkitekt import easy
from mikro import mikro_service
with easy("my-script", mikro_service) as mikro:
folder = mikro.create_folder(name="examples")
Name several services and you get a tuple back, in the same order:
from arkitekt import aeasy, interactive
from fluss import fluss_service
from mikro import mikro_service
async with aeasy("my-script", mikro_service, fluss_service) as (mikro, fluss):
...
# In Jupyter: connects once and stays connected.
mikro = interactive("notebook", mikro_service)
Every service client exposes each operation as a method, in a blocking and an a-prefixed async
flavour (mikro.create_folder(...), await mikro.acreate_folder(...)).
The CLI
arkitekt is the command line for building, running and packaging apps. Standing up an Arkitekt
server is the job of konstruktor.
| Command | What it does |
|---|---|
init |
Scaffold an app: an entrypoint file that declares it. There is no separate manifest. |
run dev · run prod |
Run the app — with hot reloading while you develop, or as it runs in a container. |
gen |
Generate typed clients. |
inspect |
Show what the app would register, without connecting. |
call |
Call an action of the app. |
plugin |
Containerize the app into flavours and publish it as a deployable plugin. |
mesh |
Join this machine to the deployment's private WireGuard mesh. |
self |
Manage your install — upgrade the SDK, print versions, dump diagnostics. |
mkdir my-app && cd my-app
arkitekt init # scaffold an app
arkitekt run dev # run it with hot reloading
See the full reference in docs/cli.md, and docs/app_types.md for choosing between a standalone and a plugin app.
Working with data
Arkitekt serializes and documents standard Python types — str, bool, int, float, Enum,
list, dict, and pydantic models or dataclasses declared with @app.model. For heavier data
(images, arrays, large objects), the platform follows a store-by-reference model: data lives in a
central, scalable store and only a lightweight reference travels between apps. Service clients like
mikro and elektro provide ready-made structures for this; within one app,
app.memory_structure(...) lets arbitrary objects cross actions without leaving the agent.
Examples
examples/ holds small, self-contained scripts — one per thing an app can do.
Each declares its dependencies in a PEP 723 header, so there is nothing to install:
uv run --script examples/hello.py
Documentation & links
- 📚 Documentation: arkitekt.live
- 🧰 CLI reference: docs/cli.md
- 📦 PyPI: pypi.org/project/arkitekt
- 🐙 Source: github.com/jhnnsrs/arkitekt
License
arkitekt is released under the MIT License.
1 maintainer
Metadata
1 maintainer
Release files for arkitekt 4.2.1
For a detailed explanation of source distributions (sdists) and built distributions (wheels), please see the package formats documentation.
Source distribution (sdist)
| File | Size | Uploaded | |
|---|---|---|---|
| arkitekt-4.2.1.tar.gz | 523.4 kB | Details |
Built distribution (wheel)
| File | Interpreter | ABI | Platform | Reset |
|---|---|---|---|---|
| arkitekt-4.2.1-py3-none-any.whl | Python 3 | none | any | Details |
Total release size: 1.1 MB
Release files / arkitekt-4.2.1.tar.gz
| Download URL | arkitekt-4.2.1.tar.gz |
|---|---|
| Size | 523.4 kB |
| Tags | Source |
|
SHA-256 checksum How to use checksums |
66fa67f6a4693481246b1b8703888c9cf06e4c8d87e09a95cd2ec2dd983397f9
|
|
BLAKE2b-256 checksum How to use checksums |
61f2155ec1dddcba51fe5ced230dcf1b2ed778d04991d3ad816efb83dc111e1f
|
| Upload date | |
|
Uploaded using Trusted Publishing? What is trusted publishing? |
Yes |
| Uploaded via |
uv/0.12.20 {"installer":{"name":"uv","version":"0.12.20","subcommand":["publish"]},"python":null,"implementation":{"name":null,"version":null},"distro":{"name":"Ubuntu","version":"24.04","id":"noble","libc":null},"system":{"name":null,"release":null},"cpu":null,"openssl_version":null,"setuptools_version":null,"rustc_version":null,"ci":true}
|
Release files / arkitekt-4.2.1-py3-none-any.whl
| Download URL | arkitekt-4.2.1-py3-none-any.whl |
|---|---|
| Size | 561.8 kB |
| Tags | Python 3 |
|
SHA-256 checksum How to use checksums |
fe6304d74fd8518e89dc1558ffc5e7b1f96bbc87ca173fc42f389d638a7ff166
|
|
BLAKE2b-256 checksum How to use checksums |
d2121c9031c56731df70db18cd2adbcb69364b854c08e7282bd296d294ae3266
|
| Upload date | |
|
Uploaded using Trusted Publishing? What is trusted publishing? |
Yes |
| Uploaded via |
uv/0.12.20 {"installer":{"name":"uv","version":"0.12.20","subcommand":["publish"]},"python":null,"implementation":{"name":null,"version":null},"distro":{"name":"Ubuntu","version":"24.04","id":"noble","libc":null},"system":{"name":null,"release":null},"cpu":null,"openssl_version":null,"setuptools_version":null,"rustc_version":null,"ci":true}
|
