openai/openai-python

Public

mirrored from https://github.com/openai/openai-pythonAvailable

CodeCommitsIssuesPull requestsActionsInsightsSecurity
v2.32.0

Branches

Tags

  • No tags available.
0Branches0Tags
Go to file
Add file
Code

Clone

HTTPS

Download ZIP

README.md

965lines · modecode

1# OpenAI Python API library
2
3<!-- prettier-ignore -->
4[![PyPI version](https://img.shields.io/pypi/v/openai.svg?label=pypi%20(stable))](https://pypi.org/project/openai/)
5
6The OpenAI Python library provides convenient access to the OpenAI REST API from any Python 3.9+
7application. The library includes type definitions for all request params and response fields,
8and offers both synchronous and asynchronous clients powered by [httpx](https://github.com/encode/httpx).
9
10It is generated from our [OpenAPI specification](https://github.com/openai/openai-openapi) with [Stainless](https://stainlessapi.com/).
11
12## Documentation
13
14The REST API documentation can be found on [platform.openai.com](https://platform.openai.com/docs/api-reference). The full API of this library can be found in [api.md](api.md).
15
16## Installation
17
18```sh
19# install from PyPI
20pip install openai
21```
22
23## Usage
24
25The full API of this library can be found in [api.md](api.md).
26
27The primary API for interacting with OpenAI models is the [Responses API](https://platform.openai.com/docs/api-reference/responses). You can generate text from the model with the code below.
28
29```python
30import os
31from openai import OpenAI
32
33client = OpenAI(
34 # This is the default and can be omitted
35 api_key=os.environ.get("OPENAI_API_KEY"),
36)
37
38response = client.responses.create(
39 model="gpt-5.2",
40 instructions="You are a coding assistant that talks like a pirate.",
41 input="How do I check if a Python object is an instance of a class?",
42)
43
44print(response.output_text)
45```
46
47The previous standard (supported indefinitely) for generating text is the [Chat Completions API](https://platform.openai.com/docs/api-reference/chat). You can use that API to generate text from the model with the code below.
48
49```python
50from openai import OpenAI
51
52client = OpenAI()
53
54completion = client.chat.completions.create(
55 model="gpt-5.2",
56 messages=[
57 {"role": "developer", "content": "Talk like a pirate."},
58 {
59 "role": "user",
60 "content": "How do I check if a Python object is an instance of a class?",
61 },
62 ],
63)
64
65print(completion.choices[0].message.content)
66```
67
68While you can provide an `api_key` keyword argument,
69we recommend using [python-dotenv](https://pypi.org/project/python-dotenv/)
70to add `OPENAI_API_KEY="My API Key"` to your `.env` file
71so that your API key is not stored in source control.
72[Get an API key here](https://platform.openai.com/settings/organization/api-keys).
73
74### Workload Identity Authentication
75
76For secure, automated environments like cloud-managed Kubernetes, Azure, and Google Cloud Platform, you can use workload identity authentication with short-lived tokens from cloud identity providers instead of long-lived API keys.
77
78#### Kubernetes (service account tokens)
79
80```python
81from openai import OpenAI
82from openai.auth import k8s_service_account_token_provider
83
84client = OpenAI(
85 workload_identity={
86 "client_id": "your-client-id",
87 "identity_provider_id": "idp-123",
88 "service_account_id": "sa-456",
89 "provider": k8s_service_account_token_provider(
90 "/var/run/secrets/kubernetes.io/serviceaccount/token"
91 ),
92 },
93 organization="org-xyz",
94 project="proj-abc",
95)
96
97response = client.chat.completions.create(
98 model="gpt-4",
99 messages=[{"role": "user", "content": "Hello!"}],
100)
101```
102
103#### Azure (managed identity)
104
105```python
106from openai import OpenAI
107from openai.auth import azure_managed_identity_token_provider
108
109client = OpenAI(
110 workload_identity={
111 "client_id": "your-client-id",
112 "identity_provider_id": "idp-123",
113 "service_account_id": "sa-456",
114 "provider": azure_managed_identity_token_provider(
115 resource="https://management.azure.com/",
116 ),
117 },
118)
119```
120
121#### Google Cloud Platform (compute engine metadata)
122
123```python
124from openai import OpenAI
125from openai.auth import gcp_id_token_provider
126
127client = OpenAI(
128 workload_identity={
129 "client_id": "your-client-id",
130 "identity_provider_id": "idp-123",
131 "service_account_id": "sa-456",
132 "provider": gcp_id_token_provider(audience="https://api.openai.com/v1"),
133 },
134)
135```
136
137#### Custom subject token provider
138
139```python
140from openai import OpenAI
141
142
143def get_custom_token() -> str:
144 return "your-jwt-token"
145
146
147client = OpenAI(
148 workload_identity={
149 "client_id": "your-client-id",
150 "identity_provider_id": "idp-123",
151 "service_account_id": "sa-456",
152 "provider": {
153 "token_type": "jwt",
154 "get_token": get_custom_token,
155 },
156 }
157)
158```
159
160You can also customize the token refresh buffer (default is 1200 seconds (20 minutes) before expiration):
161
162```python
163from openai import OpenAI
164from openai.auth import k8s_service_account_token_provider
165
166client = OpenAI(
167 workload_identity={
168 "client_id": "your-client-id",
169 "identity_provider_id": "idp-123",
170 "service_account_id": "sa-456",
171 "provider": k8s_service_account_token_provider("/var/token"),
172 "refresh_buffer_seconds": 120.0,
173 }
174)
175```
176
177### Vision
178
179With an image URL:
180
181```python
182prompt = "What is in this image?"
183img_url = "https://upload.wikimedia.org/wikipedia/commons/thumb/d/d5/2023_06_08_Raccoon1.jpg/1599px-2023_06_08_Raccoon1.jpg"
184
185response = client.responses.create(
186 model="gpt-5.2",
187 input=[
188 {
189 "role": "user",
190 "content": [
191 {"type": "input_text", "text": prompt},
192 {"type": "input_image", "image_url": f"{img_url}"},
193 ],
194 }
195 ],
196)
197```
198
199With the image as a base64 encoded string:
200
201```python
202import base64
203from openai import OpenAI
204
205client = OpenAI()
206
207prompt = "What is in this image?"
208with open("path/to/image.png", "rb") as image_file:
209 b64_image = base64.b64encode(image_file.read()).decode("utf-8")
210
211response = client.responses.create(
212 model="gpt-5.2",
213 input=[
214 {
215 "role": "user",
216 "content": [
217 {"type": "input_text", "text": prompt},
218 {"type": "input_image", "image_url": f"data:image/png;base64,{b64_image}"},
219 ],
220 }
221 ],
222)
223```
224
225## Async usage
226
227Simply import `AsyncOpenAI` instead of `OpenAI` and use `await` with each API call:
228
229```python
230import os
231import asyncio
232from openai import AsyncOpenAI
233
234client = AsyncOpenAI(
235 # This is the default and can be omitted
236 api_key=os.environ.get("OPENAI_API_KEY"),
237)
238
239
240async def main() -> None:
241 response = await client.responses.create(
242 model="gpt-5.2", input="Explain disestablishmentarianism to a smart five year old."
243 )
244 print(response.output_text)
245
246
247asyncio.run(main())
248```
249
250Functionality between the synchronous and asynchronous clients is otherwise identical.
251
252### With aiohttp
253
254By default, the async client uses `httpx` for HTTP requests. However, for improved concurrency performance you may also use `aiohttp` as the HTTP backend.
255
256You can enable this by installing `aiohttp`:
257
258```sh
259# install from PyPI
260pip install openai[aiohttp]
261```
262
263Then you can enable it by instantiating the client with `http_client=DefaultAioHttpClient()`:
264
265```python
266import os
267import asyncio
268from openai import DefaultAioHttpClient
269from openai import AsyncOpenAI
270
271
272async def main() -> None:
273 async with AsyncOpenAI(
274 api_key=os.environ.get("OPENAI_API_KEY"), # This is the default and can be omitted
275 http_client=DefaultAioHttpClient(),
276 ) as client:
277 chat_completion = await client.chat.completions.create(
278 messages=[
279 {
280 "role": "user",
281 "content": "Say this is a test",
282 }
283 ],
284 model="gpt-5.2",
285 )
286
287
288asyncio.run(main())
289```
290
291## Streaming responses
292
293We provide support for streaming responses using Server Side Events (SSE).
294
295```python
296from openai import OpenAI
297
298client = OpenAI()
299
300stream = client.responses.create(
301 model="gpt-5.2",
302 input="Write a one-sentence bedtime story about a unicorn.",
303 stream=True,
304)
305
306for event in stream:
307 print(event)
308```
309
310The async client uses the exact same interface.
311
312```python
313import asyncio
314from openai import AsyncOpenAI
315
316client = AsyncOpenAI()
317
318
319async def main():
320 stream = await client.responses.create(
321 model="gpt-5.2",
322 input="Write a one-sentence bedtime story about a unicorn.",
323 stream=True,
324 )
325
326 async for event in stream:
327 print(event)
328
329
330asyncio.run(main())
331```
332
333## Realtime API
334
335The Realtime API enables you to build low-latency, multi-modal conversational experiences. It currently supports text and audio as both input and output, as well as [function calling](https://platform.openai.com/docs/guides/function-calling) through a WebSocket connection.
336
337Under the hood the SDK uses the [`websockets`](https://websockets.readthedocs.io/en/stable/) library to manage connections.
338
339The Realtime API works through a combination of client-sent events and server-sent events. Clients can send events to do things like update session configuration or send text and audio inputs. Server events confirm when audio responses have completed, or when a text response from the model has been received. A full event reference can be found [here](https://platform.openai.com/docs/api-reference/realtime-client-events) and a guide can be found [here](https://platform.openai.com/docs/guides/realtime).
340
341Basic text based example:
342
343```py
344import asyncio
345from openai import AsyncOpenAI
346
347async def main():
348 client = AsyncOpenAI()
349
350 async with client.realtime.connect(model="gpt-realtime") as connection:
351 await connection.session.update(
352 session={"type": "realtime", "output_modalities": ["text"]}
353 )
354
355 await connection.conversation.item.create(
356 item={
357 "type": "message",
358 "role": "user",
359 "content": [{"type": "input_text", "text": "Say hello!"}],
360 }
361 )
362 await connection.response.create()
363
364 async for event in connection:
365 if event.type == "response.output_text.delta":
366 print(event.delta, flush=True, end="")
367
368 elif event.type == "response.output_text.done":
369 print()
370
371 elif event.type == "response.done":
372 break
373
374asyncio.run(main())
375```
376
377However the real magic of the Realtime API is handling audio inputs / outputs, see this example [TUI script](https://github.com/openai/openai-python/blob/main/examples/realtime/push_to_talk_app.py) for a fully fledged example.
378
379### Realtime error handling
380
381Whenever an error occurs, the Realtime API will send an [`error` event](https://platform.openai.com/docs/guides/realtime-model-capabilities#error-handling) and the connection will stay open and remain usable. This means you need to handle it yourself, as _no errors are raised directly_ by the SDK when an `error` event comes in.
382
383```py
384client = AsyncOpenAI()
385
386async with client.realtime.connect(model="gpt-realtime") as connection:
387 ...
388 async for event in connection:
389 if event.type == 'error':
390 print(event.error.type)
391 print(event.error.code)
392 print(event.error.event_id)
393 print(event.error.message)
394```
395
396## Using types
397
398Nested request parameters are [TypedDicts](https://docs.python.org/3/library/typing.html#typing.TypedDict). Responses are [Pydantic models](https://docs.pydantic.dev) which also provide helper methods for things like:
399
400- Serializing back into JSON, `model.to_json()`
401- Converting to a dictionary, `model.to_dict()`
402
403Typed requests and responses provide autocomplete and documentation within your editor. If you would like to see type errors in VS Code to help catch bugs earlier, set `python.analysis.typeCheckingMode` to `basic`.
404
405## Pagination
406
407List methods in the OpenAI API are paginated.
408
409This library provides auto-paginating iterators with each list response, so you do not have to request successive pages manually:
410
411```python
412from openai import OpenAI
413
414client = OpenAI()
415
416all_jobs = []
417# Automatically fetches more pages as needed.
418for job in client.fine_tuning.jobs.list(
419 limit=20,
420):
421 # Do something with job here
422 all_jobs.append(job)
423print(all_jobs)
424```
425
426Or, asynchronously:
427
428```python
429import asyncio
430from openai import AsyncOpenAI
431
432client = AsyncOpenAI()
433
434
435async def main() -> None:
436 all_jobs = []
437 # Iterate through items across all pages, issuing requests as needed.
438 async for job in client.fine_tuning.jobs.list(
439 limit=20,
440 ):
441 all_jobs.append(job)
442 print(all_jobs)
443
444
445asyncio.run(main())
446```
447
448Alternatively, you can use the `.has_next_page()`, `.next_page_info()`, or `.get_next_page()` methods for more granular control working with pages:
449
450```python
451first_page = await client.fine_tuning.jobs.list(
452 limit=20,
453)
454if first_page.has_next_page():
455 print(f"will fetch next page using these details: {first_page.next_page_info()}")
456 next_page = await first_page.get_next_page()
457 print(f"number of items we just fetched: {len(next_page.data)}")
458
459# Remove `await` for non-async usage.
460```
461
462Or just work directly with the returned data:
463
464```python
465first_page = await client.fine_tuning.jobs.list(
466 limit=20,
467)
468
469print(f"next page cursor: {first_page.after}") # => "next page cursor: ..."
470for job in first_page.data:
471 print(job.id)
472
473# Remove `await` for non-async usage.
474```
475
476## Nested params
477
478Nested parameters are dictionaries, typed using `TypedDict`, for example:
479
480```python
481from openai import OpenAI
482
483client = OpenAI()
484
485response = client.chat.responses.create(
486 input=[
487 {
488 "role": "user",
489 "content": "How much ?",
490 }
491 ],
492 model="gpt-5.2",
493 response_format={"type": "json_object"},
494)
495```
496
497## File uploads
498
499Request parameters that correspond to file uploads can be passed as `bytes`, or a [`PathLike`](https://docs.python.org/3/library/os.html#os.PathLike) instance or a tuple of `(filename, contents, media type)`.
500
501```python
502from pathlib import Path
503from openai import OpenAI
504
505client = OpenAI()
506
507client.files.create(
508 file=Path("input.jsonl"),
509 purpose="fine-tune",
510)
511```
512
513The async client uses the exact same interface. If you pass a [`PathLike`](https://docs.python.org/3/library/os.html#os.PathLike) instance, the file contents will be read asynchronously automatically.
514
515## Webhook Verification
516
517Verifying webhook signatures is _optional but encouraged_.
518
519For more information about webhooks, see [the API docs](https://platform.openai.com/docs/guides/webhooks).
520
521### Parsing webhook payloads
522
523For most use cases, you will likely want to verify the webhook and parse the payload at the same time. To achieve this, we provide the method `client.webhooks.unwrap()`, which parses a webhook request and verifies that it was sent by OpenAI. This method will raise an error if the signature is invalid.
524
525Note that the `body` parameter must be the raw JSON string sent from the server (do not parse it first). The `.unwrap()` method will parse this JSON for you into an event object after verifying the webhook was sent from OpenAI.
526
527```python
528from openai import OpenAI
529from flask import Flask, request
530
531app = Flask(__name__)
532client = OpenAI() # OPENAI_WEBHOOK_SECRET environment variable is used by default
533
534
535@app.route("/webhook", methods=["POST"])
536def webhook():
537 request_body = request.get_data(as_text=True)
538
539 try:
540 event = client.webhooks.unwrap(request_body, request.headers)
541
542 if event.type == "response.completed":
543 print("Response completed:", event.data)
544 elif event.type == "response.failed":
545 print("Response failed:", event.data)
546 else:
547 print("Unhandled event type:", event.type)
548
549 return "ok"
550 except Exception as e:
551 print("Invalid signature:", e)
552 return "Invalid signature", 400
553
554
555if __name__ == "__main__":
556 app.run(port=8000)
557```
558
559### Verifying webhook payloads directly
560
561In some cases, you may want to verify the webhook separately from parsing the payload. If you prefer to handle these steps separately, we provide the method `client.webhooks.verify_signature()` to _only verify_ the signature of a webhook request. Like `.unwrap()`, this method will raise an error if the signature is invalid.
562
563Note that the `body` parameter must be the raw JSON string sent from the server (do not parse it first). You will then need to parse the body after verifying the signature.
564
565```python
566import json
567from openai import OpenAI
568from flask import Flask, request
569
570app = Flask(__name__)
571client = OpenAI() # OPENAI_WEBHOOK_SECRET environment variable is used by default
572
573
574@app.route("/webhook", methods=["POST"])
575def webhook():
576 request_body = request.get_data(as_text=True)
577
578 try:
579 client.webhooks.verify_signature(request_body, request.headers)
580
581 # Parse the body after verification
582 event = json.loads(request_body)
583 print("Verified event:", event)
584
585 return "ok"
586 except Exception as e:
587 print("Invalid signature:", e)
588 return "Invalid signature", 400
589
590
591if __name__ == "__main__":
592 app.run(port=8000)
593```
594
595## Handling errors
596
597When the library is unable to connect to the API (for example, due to network connection problems or a timeout), a subclass of `openai.APIConnectionError` is raised.
598
599When the API returns a non-success status code (that is, 4xx or 5xx
600response), a subclass of `openai.APIStatusError` is raised, containing `status_code` and `response` properties.
601
602All errors inherit from `openai.APIError`.
603
604```python
605import openai
606from openai import OpenAI
607
608client = OpenAI()
609
610try:
611 client.fine_tuning.jobs.create(
612 model="gpt-4o",
613 training_file="file-abc123",
614 )
615except openai.APIConnectionError as e:
616 print("The server could not be reached")
617 print(e.__cause__) # an underlying Exception, likely raised within httpx.
618except openai.RateLimitError as e:
619 print("A 429 status code was received; we should back off a bit.")
620except openai.APIStatusError as e:
621 print("Another non-200-range status code was received")
622 print(e.status_code)
623 print(e.response)
624```
625
626Error codes are as follows:
627
628| Status Code | Error Type |
629| ----------- | -------------------------- |
630| 400 | `BadRequestError` |
631| 401 | `AuthenticationError` |
632| 403 | `PermissionDeniedError` |
633| 404 | `NotFoundError` |
634| 422 | `UnprocessableEntityError` |
635| 429 | `RateLimitError` |
636| >=500 | `InternalServerError` |
637| N/A | `APIConnectionError` |
638
639## Request IDs
640
641> For more information on debugging requests, see [these docs](https://platform.openai.com/docs/api-reference/debugging-requests)
642
643All object responses in the SDK provide a `_request_id` property which is added from the `x-request-id` response header so that you can quickly log failing requests and report them back to OpenAI.
644
645```python
646response = await client.responses.create(
647 model="gpt-5.2",
648 input="Say 'this is a test'.",
649)
650print(response._request_id) # req_123
651```
652
653Note that unlike other properties that use an `_` prefix, the `_request_id` property
654_is_ public. Unless documented otherwise, _all_ other `_` prefix properties,
655methods and modules are _private_.
656
657> [!IMPORTANT]
658> If you need to access request IDs for failed requests you must catch the `APIStatusError` exception
659
660```python
661import openai
662
663try:
664 completion = await client.chat.completions.create(
665 messages=[{"role": "user", "content": "Say this is a test"}], model="gpt-5.2"
666 )
667except openai.APIStatusError as exc:
668 print(exc.request_id) # req_123
669 raise exc
670```
671
672## Retries
673
674Certain errors are automatically retried 2 times by default, with a short exponential backoff.
675Connection errors (for example, due to a network connectivity problem), 408 Request Timeout, 409 Conflict,
676429 Rate Limit, and >=500 Internal errors are all retried by default.
677
678You can use the `max_retries` option to configure or disable retry settings:
679
680```python
681from openai import OpenAI
682
683# Configure the default for all requests:
684client = OpenAI(
685 # default is 2
686 max_retries=0,
687)
688
689# Or, configure per-request:
690client.with_options(max_retries=5).chat.completions.create(
691 messages=[
692 {
693 "role": "user",
694 "content": "How can I get the name of the current day in JavaScript?",
695 }
696 ],
697 model="gpt-5.2",
698)
699```
700
701## Timeouts
702
703By default requests time out after 10 minutes. You can configure this with a `timeout` option,
704which accepts a float or an [`httpx.Timeout`](https://www.python-httpx.org/advanced/timeouts/#fine-tuning-the-configuration) object:
705
706```python
707from openai import OpenAI
708
709# Configure the default for all requests:
710client = OpenAI(
711 # 20 seconds (default is 10 minutes)
712 timeout=20.0,
713)
714
715# More granular control:
716client = OpenAI(
717 timeout=httpx.Timeout(60.0, read=5.0, write=10.0, connect=2.0),
718)
719
720# Override per-request:
721client.with_options(timeout=5.0).chat.completions.create(
722 messages=[
723 {
724 "role": "user",
725 "content": "How can I list all files in a directory using Python?",
726 }
727 ],
728 model="gpt-5.2",
729)
730```
731
732On timeout, an `APITimeoutError` is thrown.
733
734Note that requests that time out are [retried twice by default](#retries).
735
736## Advanced
737
738### Logging
739
740We use the standard library [`logging`](https://docs.python.org/3/library/logging.html) module.
741
742You can enable logging by setting the environment variable `OPENAI_LOG` to `info`.
743
744```shell
745$ export OPENAI_LOG=info
746```
747
748Or to `debug` for more verbose logging.
749
750### How to tell whether `None` means `null` or missing
751
752In an API response, a field may be explicitly `null`, or missing entirely; in either case, its value is `None` in this library. You can differentiate the two cases with `.model_fields_set`:
753
754```py
755if response.my_field is None:
756 if 'my_field' not in response.model_fields_set:
757 print('Got json like {}, without a "my_field" key present at all.')
758 else:
759 print('Got json like {"my_field": null}.')
760```
761
762### Accessing raw response data (e.g. headers)
763
764The "raw" Response object can be accessed by prefixing `.with_raw_response.` to any HTTP method call, e.g.,
765
766```py
767from openai import OpenAI
768
769client = OpenAI()
770response = client.chat.completions.with_raw_response.create(
771 messages=[{
772 "role": "user",
773 "content": "Say this is a test",
774 }],
775 model="gpt-5.2",
776)
777print(response.headers.get('X-My-Header'))
778
779completion = response.parse() # get the object that `chat.completions.create()` would have returned
780print(completion)
781```
782
783These methods return a [`LegacyAPIResponse`](https://github.com/openai/openai-python/tree/main/src/openai/_legacy_response.py) object. This is a legacy class as we're changing it slightly in the next major version.
784
785For the sync client this will mostly be the same with the exception
786of `content` & `text` will be methods instead of properties. In the
787async client, all methods will be async.
788
789A migration script will be provided & the migration in general should
790be smooth.
791
792#### `.with_streaming_response`
793
794The above interface eagerly reads the full response body when you make the request, which may not always be what you want.
795
796To stream the response body, use `.with_streaming_response` instead, which requires a context manager and only reads the response body once you call `.read()`, `.text()`, `.json()`, `.iter_bytes()`, `.iter_text()`, `.iter_lines()` or `.parse()`. In the async client, these are async methods.
797
798As such, `.with_streaming_response` methods return a different [`APIResponse`](https://github.com/openai/openai-python/tree/main/src/openai/_response.py) object, and the async client returns an [`AsyncAPIResponse`](https://github.com/openai/openai-python/tree/main/src/openai/_response.py) object.
799
800```python
801with client.chat.completions.with_streaming_response.create(
802 messages=[
803 {
804 "role": "user",
805 "content": "Say this is a test",
806 }
807 ],
808 model="gpt-5.2",
809) as response:
810 print(response.headers.get("X-My-Header"))
811
812 for line in response.iter_lines():
813 print(line)
814```
815
816The context manager is required so that the response will reliably be closed.
817
818### Making custom/undocumented requests
819
820This library is typed for convenient access to the documented API.
821
822If you need to access undocumented endpoints, params, or response properties, the library can still be used.
823
824#### Undocumented endpoints
825
826To make requests to undocumented endpoints, you can make requests using `client.get`, `client.post`, and other
827http verbs. Options on the client will be respected (such as retries) when making this request.
828
829```py
830import httpx
831
832response = client.post(
833 "/foo",
834 cast_to=httpx.Response,
835 body={"my_param": True},
836)
837
838print(response.headers.get("x-foo"))
839```
840
841#### Undocumented request params
842
843If you want to explicitly send an extra param, you can do so with the `extra_query`, `extra_body`, and `extra_headers` request
844options.
845
846#### Undocumented response properties
847
848To access undocumented response properties, you can access the extra fields like `response.unknown_prop`. You
849can also get all the extra fields on the Pydantic model as a dict with
850[`response.model_extra`](https://docs.pydantic.dev/latest/api/base_model/#pydantic.BaseModel.model_extra).
851
852### Configuring the HTTP client
853
854You can directly override the [httpx client](https://www.python-httpx.org/api/#client) to customize it for your use case, including:
855
856- Support for [proxies](https://www.python-httpx.org/advanced/proxies/)
857- Custom [transports](https://www.python-httpx.org/advanced/transports/)
858- Additional [advanced](https://www.python-httpx.org/advanced/clients/) functionality
859
860```python
861import httpx
862from openai import OpenAI, DefaultHttpxClient
863
864client = OpenAI(
865 # Or use the `OPENAI_BASE_URL` env var
866 base_url="http://my.test.server.example.com:8083/v1",
867 http_client=DefaultHttpxClient(
868 proxy="http://my.test.proxy.example.com",
869 transport=httpx.HTTPTransport(local_address="0.0.0.0"),
870 ),
871)
872```
873
874You can also customize the client on a per-request basis by using `with_options()`:
875
876```python
877client.with_options(http_client=DefaultHttpxClient(...))
878```
879
880### Managing HTTP resources
881
882By default the library closes underlying HTTP connections whenever the client is [garbage collected](https://docs.python.org/3/reference/datamodel.html#object.__del__). You can manually close the client using the `.close()` method if desired, or with a context manager that closes when exiting.
883
884```py
885from openai import OpenAI
886
887with OpenAI() as client:
888 # make requests here
889 ...
890
891# HTTP client is now closed
892```
893
894## Microsoft Azure OpenAI
895
896To use this library with [Azure OpenAI](https://learn.microsoft.com/azure/ai-services/openai/overview), use the `AzureOpenAI`
897class instead of the `OpenAI` class.
898
899> [!IMPORTANT]
900> The Azure API shape differs from the core API shape which means that the static types for responses / params
901> won't always be correct.
902
903```py
904from openai import AzureOpenAI
905
906# gets the API Key from environment variable AZURE_OPENAI_API_KEY
907client = AzureOpenAI(
908 # https://learn.microsoft.com/azure/ai-services/openai/reference#rest-api-versioning
909 api_version="2023-07-01-preview",
910 # https://learn.microsoft.com/azure/cognitive-services/openai/how-to/create-resource?pivots=web-portal#create-a-resource
911 azure_endpoint="https://example-endpoint.openai.azure.com",
912)
913
914completion = client.chat.completions.create(
915 model="deployment-name", # e.g. gpt-35-instant
916 messages=[
917 {
918 "role": "user",
919 "content": "How do I output all files in a directory using Python?",
920 },
921 ],
922)
923print(completion.to_json())
924```
925
926In addition to the options provided in the base `OpenAI` client, the following options are provided:
927
928- `azure_endpoint` (or the `AZURE_OPENAI_ENDPOINT` environment variable)
929- `azure_deployment`
930- `api_version` (or the `OPENAI_API_VERSION` environment variable)
931- `azure_ad_token` (or the `AZURE_OPENAI_AD_TOKEN` environment variable)
932- `azure_ad_token_provider`
933
934An example of using the client with Microsoft Entra ID (formerly known as Azure Active Directory) can be found [here](https://github.com/openai/openai-python/blob/main/examples/azure_ad.py).
935
936## Versioning
937
938This package generally follows [SemVer](https://semver.org/spec/v2.0.0.html) conventions, though certain backwards-incompatible changes may be released as minor versions:
939
9401. Changes that only affect static types, without breaking runtime behavior.
9412. Changes to library internals which are technically public but not intended or documented for external use. _(Please open a GitHub issue to let us know if you are relying on such internals.)_
9423. Changes that we do not expect to impact the vast majority of users in practice.
943
944We take backwards-compatibility seriously and work hard to ensure you can rely on a smooth upgrade experience.
945
946We are keen for your feedback; please open an [issue](https://www.github.com/openai/openai-python/issues) with questions, bugs, or suggestions.
947
948### Determining the installed version
949
950If you've upgraded to the latest version but aren't seeing any new features you were expecting then your python environment is likely still using an older version.
951
952You can determine the version that is being used at runtime with:
953
954```py
955import openai
956print(openai.__version__)
957```
958
959## Requirements
960
961Python 3.9 or higher.
962
963## Contributing
964
965See [the contributing documentation](./CONTRIBUTING.md).
966