fastsdk 0.2.33

Your SDK and model zoo for generative AI. Build AI-powered applications with ease.

Turn any hosted service into a typed Python client.

Point fastSDK at an OpenAPI spec and get a Python client with a method per endpoint,
typed parameters, file uploads, and job handling. Call web APIs like plain functions.

Why

You built a web service with FastAPI, Flask, Cog, or APIPod. Now you need a Python client.

Hand-writing one with requests works until it does not. You add endpoints. You run calls in parallel. You transfer a 1GB video and the request times out. You fall into threading and asyncio complexity, and the client drifts out of sync with the API.

fastSDK generates the client from your spec. Endpoints become typed methods. Files upload and stream. Long-running calls return a job you can wait on or cancel. The heavy I/O runs concurrently under the hood through meseex, so your code stays synchronous and simple.

Install

pip install fastsdk

Quick start

Grab your service's openapi.json (usually at http://localhost:8000/openapi.json), then generate a client:

from fastsdk import FastSDK

# generate a client file from the spec
FastSDK().create_sdk("openapi.json", save_path="my_service.py", class_name="MyService")

The generated file has one method per endpoint, with every parameter and default. Import it and call it:

from my_service import MyService

client = MyService(api_key="my_api_key")
job = client.my_method(...)          # returns a job immediately, does not block
result = job.wait_for_result()       # typed result back

Need a one-off client without writing a file? Skip generation:

from fastsdk import FastSDK

client = FastSDK().create_temporary_client("openapi.json", api_key="my_api_key")

Jobs

Every endpoint call returns a job. Start many, collect them later.

job = client.my_method(...)

job.wait_for_result()   # block until done, return the typed result
job.cancel()            # cancel locally if queued, or ask the provider to cancel

For job-based providers (APIPod, Runpod, Socaity, Replicate), fastSDK polls remote status and reports progress until the job reaches a terminal state.

Files

File parameters accept paths or media-toolkit objects. Large files upload to cloud storage (Azure Blob, Amazon S3) and are passed by URL; results come back as typed media objects you can save.

from fastsdk import ImageFile

job = client.swap_img_to_img("face_1.jpg", "face_2.jpg")
result = job.wait_for_result()
result.save("swapped.jpg")

MediaFile, ImageFile, VideoFile, and AudioFile are importable straight from the package.

API keys

Pass the key when you create the client, or read it from the environment:

import os
from my_service import MyService

client = MyService(api_key=os.getenv("MY_API_KEY"))

Service compatibility

Works out of the box with:

Provider	Notes
OpenAPI 3.0 / REST	FastAPI, Flask, any compliant spec
APIPod	job-based services
Runpod	serverless GPU endpoints
Cog	Replicate-style services
Socaity	hosted models
Replicate	prediction APIs

fastSDK and APIPod

APIPod builds and deploys the services. fastSDK calls them. Two halves of the same client-to-service loop, designed for long-running ML and data workloads.

Architecture

For internals (definition layer, runtime pipeline, cancellation, how meseex fits in) see TECHNICAL_README.md.

Status

Alpha. Syntax and surface change rapidly. Bug reports, ideas, and pull requests are welcome in the issues section.

fastSDK is licensed under GPLv3 and free to use. Leave a star to support us.