RunAPI Core Python SDK

The RunAPI Core Python SDK provides shared authentication, HTTP, retry, error, Files, Uploads, and polling primitives for RunAPI model packages. Install runapi-core only when you are building SDK infrastructure or shared Python tooling; application code should normally install a concrete model package such as runapi-flux-2.

Install

pip install runapi-core

Notes

Use the core package for common client options, error classes, request helpers, file uploads, live pricing, account resources, and task polling behavior that Provider Clients share. Configure it globally or per client:

import runapi.core as runapi

runapi.configure(api_key="sk-...")  # or set RUNAPI_API_KEY in the environment

Request identifiers

RunAPI accepts an optional X-Client-Request-Id header on public API calls. Use printable ASCII values up to 512 characters. Accepted values are echoed in the response and stored with the RunAPI task for support and reconciliation.

Task-creation calls also accept an optional opaque Idempotency-Key up to 512 characters. Generate one value per logical task and reuse it only with identical input after an unknown result. Reusing the value with different input returns 409 Conflict; do not derive it from X-Client-Request-Id.

Recovering Interrupted Requests

Normal run() calls wait for and return the endpoint's terminal result. It also generates an Idempotency-Key before the initial POST when one was not supplied. If a lower-level request is interrupted after RunAPI accepted it, persist the opaque Location from the response and call subscribe(location) to recover the terminal result. Completed JSON responses keep their endpoint-specific type; text, SRT, and VTT responses return str.

from runapi.midjourney import MidjourneyClient

client = MidjourneyClient(api_key="sk-...")
result = client.shorten_prompt.run(prompt="A detailed cinematic mountain landscape")

# Persist this opaque URL after receiving 202 from a lower-level request.
result = client.shorten_prompt.subscribe("https://runapi.ai/api/v1/tasks/task-id")

High-level Python Provider Client resource methods accept per-request options and keep response headers on the returned model object. This example uses the Suno Provider Client; install runapi-suno to run it.

import os

from runapi.core import RequestOptions
from runapi.suno import SunoClient

client = SunoClient(api_key=os.environ["RUNAPI_API_KEY"])
options = RequestOptions(
    headers={
        "X-Client-Request-Id": "order-123",
        "Idempotency-Key": "opaque-logical-task-123",
    },
)

response = client.text_to_music.create(
    prompt="A chill lo-fi beat",
    model="suno-v4.5-plus",
    vocal_mode="instrumental",
    options=options,
)

runapi_task_id = response.runapi_task_id
# Equivalent case-insensitive lookup:
runapi_task_id = response.response_headers["X-RunAPI-Task-Id"]

from runapi.core import FilesClient

files = FilesClient()  # reads RUNAPI_API_KEY, or pass api_key="sk-..."
uploaded = files.create(file="./input.png")
remote = files.create(source="https://cdn.runapi.ai/public/samples/input.png")
inline = files.create(source="iVBORw0KGgo...")
print(uploaded.url)

The existing files.create() method keeps its temporary URL behavior. Every Provider Client also exposes persistent Files and multipart Uploads:

file = client.files.create_file(file="./knowledge.pdf", purpose="user_data")
content = client.files.content(file.id)

upload = client.uploads.create(
    bytes=1048576,
    filename="archive.bin",
    mime_type="application/octet-stream",
)
part = client.uploads.add_part(upload.id, data="./archive.part-01")
completed = client.uploads.complete(upload.id, part_ids=[part.id])

Use files.list(), retrieve(), and delete_file() for the remaining File lifecycle. See https://runapi.ai/docs/resources/files for limits and REST examples.

Public SDK docs live at https://runapi.ai/docs/resources/sdks and the model catalog lives at https://runapi.ai/models.

Universal resources

Every Provider Client exposes files, account, and pricing through one shared HTTP client. pricing.list_schedules() reads the current schedule, and pricing.create_quote() estimates a reservation without creating a Task. The schedule and ordinary quote calls do not require an API key; a quote that uses an existing Task as pricing context requires the key for that Task's Account.

from runapi.core import RequestOptions
from runapi.flux import FluxClient

client = FluxClient(api_key="sk-...")
schedules = client.pricing.list_schedules(
    service="flux",
    action="text_to_image",
    options=RequestOptions(headers={"If-None-Match": '"previous-etag"'}),
)
quote = client.pricing.create_quote(
    service="flux",
    action="text_to_image",
    model="flux-pro",
    params={"prompt": "A glass observatory"},
)

Completed Task query and webhook envelopes expose usage.cost as a float USD amount for that Task. Create, processing, and failed envelopes omit usage. The value is the recorded settlement for that Task and is never recalculated from the current price schedule.

License

Licensed under the Apache License, Version 2.0.