Call any AI / web service like a native Python function
fastSDK turns any hosted service — OpenAPI/FastAPI, APIPod, RunPod, Replicate, Cog — into a Python client that feels like a local library: typed methods, file upload/download, async job handling and parallel execution included.
Point it at a service. Call it like a function. That's the whole idea.
import fastsdk
client = fastsdk.connect("http://localhost:8009")
job = client.submit_job("/text2voice", text="hello world")
audio = job.get_result()
audio.save("hello.mp3")
Why fastSDK?
Calling a web service from Python sounds trivial until you actually do it in production:
you wait synchronously on long-running ML jobs, you hand-write request code for every endpoint, you fight with file uploads (try sending a 1 GB video through requests), you poll job status loops, and you reinvent threading to run requests in parallel.
fastSDK solves exactly that, and nothing else:
- One call = one job. Every call returns a job object immediately. Get the result when you need it, run hundreds of jobs in parallel meanwhile.
- Files just work. Images, audio, video are handled by media-toolkit — local paths, URLs, bytes or numpy arrays in; media objects out. Large files are uploaded via cloud storage (S3, Azure) when configured.
- Job-based providers are normalized. Replicate, RunPod serverless, APIPod and Socaity all expose "submit, then poll" APIs with different wire formats. fastSDK handles submission, polling, progress and cancellation uniformly.
- Codegen when you want it, not when you don't. Use
connect()for instant access, orgenerate_stub()to get a typed.pyclient with one method per endpoint - autocomplete and docstrings included.
Installation
pip install fastsdk # core
pip install fastsdk[replicate] # + Replicate model support
Get started
Option A: connect — use a service right now
No files, no codegen. Works with a URL, an openapi.json path, or a Replicate model reference.
import fastsdk
client = fastsdk.connect("http://localhost:8009")
job = client.submit_job("/text2voice", text="hello world")
result = job.get_result()
Option B: generate_stub — typed clients for real projects
Generates a .py file with one typed method per endpoint. This is your SDK.
import fastsdk
stub = fastsdk.generate_stub("http://localhost:8009", save_path="clients/")
print(stub.path, stub.class_name)
# use it immediately ...
client = stub.client()
job = client.text2voice(text="hello world")
# ... or import it in your next run like any other module
# from clients.speechcraft import speechcraft
# client = speechcraft()
Re-running generate_stub is safe: the file is overwritten and the service registration is updated, not duplicated.
Replicate models
Official models (called via /v1/models/{owner}/{name}/predictions) and community models (called via /v1/predictions with a version) are resolved automatically — you just name the model:
import fastsdk # requires: pip install fastsdk[replicate] and REPLICATE_API_KEY
stub = fastsdk.generate_stub("replicate:black-forest-labs/flux-schnell", save_path="clients/")
flux = stub.client()
job = flux(prompt="a t-rex on a skateboard")
image = job.get_result()
Working with jobs
job = client.swap_img_to_img(source_img="face1.jpg", target_img="face2.jpg")
job.get_result() # block until done and return the result
job.cancel() # cancel locally and remotely (provider permitting)
# run many jobs in parallel - this is where fastSDK shines
jobs = [client.text2voice(text=t) for t in hundred_texts]
results = fastsdk.gather_results(jobs)
API keys
Pass api_key=... to connect(), generate_stub() or the client constructor — or set environment variables: REPLICATE_API_KEY, RUNPOD_API_KEY, SOCAITY_API_KEY, or <SERVICE_ID>_API_KEY for your own services.
The four concepts
| Concept | What it is |
|---|---|
| Service | An AIService (from socaity-schemas) with one deployment: hosting provider, address, and the parsed ServiceContract (endpoints, parameters, whether responses are polled jobs). Get one with fastsdk.inspect_service(source), it has no side effects. |
| Registry | An in-process directory of services, shared by all clients. register_service() adds to it; generated stubs look their service up in it by ID. |
| Client | The runtime object you call (FastClient). It submits jobs to the service. connect() gives you a generic one instantly. |
| Stub | A generated .py file containing a client subclass with one typed method per endpoint. Made by generate_stub(); it's plain code — read it, version it, ship it. |
CLI
Everything above also works from the terminal — same verbs, same behavior:
# What can this service do?
fastsdk inspect http://localhost:8009
# Generate a typed client stub
fastsdk generate http://localhost:8009 -o clients/ --name SpeechCraft
fastsdk generate replicate:black-forest-labs/flux-schnell --api-key r8_...
# Call an endpoint without writing any code (curl for AI services)
fastsdk call http://localhost:8009 /text2voice --text "hello world" -o hello.mp3
# Keep services around by name (persisted in ~/.fastsdk/registry)
fastsdk registry add http://localhost:8009 --name speechcraft
fastsdk registry list
fastsdk call speechcraft /text2voice --text "hi again"
Service compatibility
Works out of the box with:
- APIPod services (job-based, the natural counterpart to fastSDK)
- Replicate models (official and community)
- RunPod serverless endpoints
- Cog services
- Any OpenAPI 3.0 service (FastAPI, Flask, ...)
- Socaity.ai services
fastSDK + APIPod
APIPod builds and deploys the services; fastSDK consumes them. Two beating hearts :two_hearts: for service ↔ client interaction.
Contribute
We at SocAIty want to provide the best tools to bring generative AI to the cloud. Report bugs, ideas and feature requests in the issues section. fastSDK is MIT-licensed and free to use. Leave a star to support us!
Made with ❤️ by SocAIty
1 maintainer
Download files
Download the file for your platform. If you're not sure which to choose, learn more about installing packages.
Source Distribution
Built Distribution
Filter files by name, interpreter, ABI, and platform.
If you're not sure about the file name format, learn more about wheel file names.
Copy a direct link to the current filters
File details
Details for the file fastsdk-0.4.15.tar.gz.
File metadata
- Download URL: fastsdk-0.4.15.tar.gz
- Upload date:
- Size: 100.6 kB
- Tags: Source
- Uploaded using Trusted Publishing? No
- Uploaded via:
twine/7.0.0 CPython/3.12.13
File hashes
| Algorithm | Hash digest | |
|---|---|---|
| SHA256 |
67826ad36cbfb75e4c93787011012b4a6ad8eadb331561d670582a3a96584aa9
|
|
| MD5 |
51fe2f4f1cf67f9f058294fdf41716ea
|
|
| BLAKE2b-256 |
2eb49af2715635ff44535dea1741411cbd64cd7e8309813527dd42c4cc15c8cc
|
File details
Details for the file fastsdk-0.4.15-py3-none-any.whl.
File metadata
- Download URL: fastsdk-0.4.15-py3-none-any.whl
- Upload date:
- Size: 88.7 kB
- Tags: Python 3
- Uploaded using Trusted Publishing? No
- Uploaded via:
twine/7.0.0 CPython/3.12.13
File hashes
| Algorithm | Hash digest | |
|---|---|---|
| SHA256 |
829a0cc1543f066df353876dc256f7ae2885db509cde1b5ed96ca42f0b8dd970
|
|
| MD5 |
73b8441eaf36cde783d1b9d34f253c01
|
|
| BLAKE2b-256 |
ee6d497980803bd46b554b2ff9bc4d28f6027a1619d4d846f74a1790e86bc24a
|
