使用后台模式可以异步执行耗时较长的 Responses API 请求,包括多步骤智能体运行中的请求。一次智能体运行涵盖由应用管理的整个工作流。应用先发送初始请求,再根据结果决定下一步,例如运行工具并在后续请求中发送工具输出。如果多个请求彼此独立,而且一开始就可以一起提交,请使用 Batch API。
根据您需要多快收到初始响应来设置 background_options.completion_window。选择 2h 或 8h 时,API 的目标是在接受请求后的相应时限内完成响应。
| 完成时限 |
适用场景 |
模型 Token 折扣 |
soonest(默认) |
有人正在等待结果时,按需生成报告。 |
无 |
2h |
在下午的分流会议前分析上午收到的支持工单。 |
比标准费率低 30% |
8h |
夜间审查一批文档,供次日早晨使用。 |
比标准费率低 40% |
如需在夜间审查文档,先发送一个 8h 请求。如果初始响应要求应用查找文档,请运行工具,并在后续请求中发送工具结果。沿用相同的完成时限、续接 ID、模型和项目,使后续请求有资格获得更高的队列优先级。
目标完成时限从 API 接受初始请求时开始计算,仅适用于初始响应。每个符合条件的请求(包括后续请求)均可享受模型 Token 费率折扣。在此示例中,两个请求均享受 40% 的折扣。
将 background 设为 true,并保存 API 返回的响应 ID。通过 webhook 获取请求完成通知,再按 ID 获取响应。本示例使用默认完成时限(soonest)。
1
2
3
4
5
6
7
8curl https://api.openai.com/v1/responses \
-H "Content-Type: application/json" \
-H "Authorization: Bearer $OPENAI_API_KEY" \
-d '{
"model": "gpt-6.2-astra",
"input": "Write a very long novel about otters in space.",
"background": true
}'
1
2
3
4
5
6
7
8
9
10import OpenAI from "openai";
const client = new OpenAI();
const resp = await client.responses.create({
model: "gpt-6.2-astra",
input: "Write a very long novel about otters in space.",
background: true,
});
console.log(resp.status);
1
2
3
4
5
6
7
8
9
10
11from openai import OpenAI
client = OpenAI()
resp = client.responses.create(
model="gpt-6.2-astra",
input="Write a very long novel about otters in space.",
background=True,
)
print(resp.status)
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26package main
import (
"context"
"fmt"
"github.com/openai/openai-go/v3"
"github.com/openai/openai-go/v3/responses"
)
func main() {
client := openai.NewClient()
response, err := client.Responses.New(context.Background(), responses.ResponseNewParams{
Model: "gpt-6.2-astra",
Background: openai.Bool(true),
Input: responses.ResponseNewParamsInputUnion{
OfString: openai.String("Write a very long novel about otters in space."),
},
})
if err != nil {
panic(err)
}
fmt.Println(response.Status)
}
1
2
3
4
5
6
7
8
9
10
11
12
13import com.openai.client.OpenAIClient;
import com.openai.client.okhttp.OpenAIOkHttpClient;
import com.openai.models.responses.ResponseCreateParams;
ResponseCreateParams params =
ResponseCreateParams.builder()
.model("gpt-6.2-astra")
.input("Write a detailed market analysis.")
.background(true)
.build();
var response = client.responses().create(params);
System.out.println(response.status().orElseThrow());
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17using OpenAI.Responses;
#pragma warning disable OPENAI001
string key = Environment.GetEnvironmentVariable("OPENAI_API_KEY")!;
ResponsesClient client = new(key);
CreateResponseOptions options = new()
{
Model = "gpt-6.2-astra",
BackgroundModeEnabled = true,
};
options.InputItems.Add(
ResponseItem.CreateUserMessageItem("Write a very long novel about otters in space.")
);
ResponseResult response = await client.CreateResponseAsync(options);
Console.WriteLine(response.Status);
1
2
3
4
5
6
7
8
9
10require "openai"
client = OpenAI::Client.new
response = client.responses.create(
model: "gpt-6.2-astra",
input: "Write a detailed market analysis.",
background: true
)
puts(response.status)
即使将 store 设为 false 或使用零数据保留,后台模式仍会临时存储响应数据。API 使用这些数据来执行请求,并提供结果供后续获取。对于所有完成时限(soonest、2h 和 8h),API 都会对临时存储的请求内容进行静态加密。有关临时存储的详情,请参阅数据控制。
- 为发送后台请求的项目创建 webhook 端点。订阅
response.completed,并按需订阅与失败、未完成或已取消的响应对应的事件。
- 对收到的事件验证签名。将耗时较长的工作交给后台工作进程,接受事件后立即返回
2xx 响应。
- 对于
response.completed 事件,使用 event.data.id 获取响应。其他终态事件按应用的工作流处理。
Webhook 指南提供了一个验证签名并获取已完成响应的服务端示例。如果无法接收 webhook,请使用响应 ID 轮询响应。
API 接受 2h 或 8h 请求后,如果遇到可重试的容量错误,会自动重试。当响应处于 queued 或 in_progress 状态时,请等待 webhook,或使用同一个响应 ID 轮询。
从 GPT-6.2 Astra 和 GPT-6.2 Sol 开始,新模型在发布时不提供 Flex 处理(service_tier: flex)。对于耗时较长的 Responses API 请求,请在支持此功能的模型上使用标准服务层级,并选择享受折扣的完成时限。
享受折扣的完成时限目前在 GPT-6.2 Astra、GPT-6.2 Sol、GPT-6.1 Terra 和 GPT-6 Luna 上开放公测。
完成时限为 2h 或 8h 的请求与 Batch API 作业共用批处理与后台请求的速率限制。API 会在请求提交时统计输入 Token 数量。使用默认完成时限 soonest 的请求适用常规请求数和 Token 数限制。
如果应用可以等待初始响应,请使用 2 小时或 8 小时的完成时限。将 background 设为 true,并将 background_options.completion_window 设为 2h 或 8h。省略 stream 或将其设为 false。请参阅完成时限定价。
享受完成时限折扣的请求必须按标准服务层级处理。将 service_tier 设为 default 即可选择标准服务层级。使用 auto 时,只有 API 选择标准服务层级,请求才符合折扣条件。
即使提示词完全相同,未使用折扣时限的标准请求也无法复用由 2h 或 8h 后台请求创建的提示词缓存条目。
如需通过一次请求在夜间审查文档,请发送 8h 请求至 POST /v1/responses,使用 gpt-6.1-terra,并在 input 中传入文档内容。如果只需要一个响应,请省略 background_options.continuation_id。
如需使用默认完成时限,请省略 background_options.completion_window 或将其设为 soonest。
在夜间文档审查流程中,两次请求应使用同一个 API 项目。
- 要求
gpt-6.1-terra 按 ID 查找文档,并检查文档是否缺少所需审批。请在创建响应的请求中提出这一要求。将 background 设为 true,将 service_tier 设为 default,并将 background_options.completion_window 设为 8h。将 background_options.continuation_id 设为应用为本次运行指定的唯一 ID。在请求中提供一个可按 ID 获取文档的函数工具。保存 API 返回的响应 ID。
- 收到
response.completed webhook 时,请对重复投递的事件进行去重,去重依据是 webhook-id 请求头。确认 event.data.id 与保存的响应 ID 一致后,再继续处理。获取响应,并按其中的 function_call 输出项执行文档查找。如果应用可能需要在响应过期后继续运行,请在仍可获取响应时保存原始输入和完整的响应 output。
- 创建后续响应,沿用相同的模型、服务层级、后台设置、完成时限和续接 ID。如果上一条响应仍可获取,请将
previous_response_id 设为 event.data.id,以延续对话历史。在 input 中发送一个 function_call_output 项,使用原函数调用的 call_id,并将获取到的文档作为该项的 output。
复用 continuation_id,使后续请求有资格获得更高的队列优先级。实际开始时间取决于调度。如果上一条响应已过期,请省略 previous_response_id,并在 input 中传入对话历史。按原顺序重新传入原始输入和已保存的响应 output 中的每一项,包括其中的所有加密推理项和原函数调用。然后追加新的 function_call_output 项。
除了使用 webhook,您也可以使用响应 ID 调用响应获取端点,以查看请求状态。状态为 queued 或 in_progress 时,请继续轮询。进入终态后,停止轮询。
1
2
3curl https://api.openai.com/v1/responses/resp_123 \
-H "Content-Type: application/json" \
-H "Authorization: Bearer $OPENAI_API_KEY"
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16import OpenAI from "openai";
const client = new OpenAI();
let resp = await client.responses.create({
model: "gpt-6.2-astra",
input: "Write a very long novel about otters in space.",
background: true,
});
while (resp.status === "queued" || resp.status === "in_progress") {
console.log("Current status: " + resp.status);
await new Promise((resolve) => setTimeout(resolve, 2000)); // wait 2 seconds
resp = await client.responses.retrieve(resp.id);
}
console.log("Final status: " + resp.status + "\nOutput:\n" + resp.output_text);
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17from openai import OpenAI
from time import sleep
client = OpenAI()
resp = client.responses.create(
model="gpt-6.2-astra",
input="Write a very long novel about otters in space.",
background=True,
)
while resp.status in {"queued", "in_progress"}:
print(f"Current status: {resp.status}")
sleep(2)
resp = client.responses.retrieve(resp.id)
print(f"Final status: {resp.status}\nOutput:\n{resp.output_text}")
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36package main
import (
"context"
"fmt"
"time"
"github.com/openai/openai-go/v3"
"github.com/openai/openai-go/v3/responses"
)
func main() {
client := openai.NewClient()
response, err := client.Responses.New(context.Background(), responses.ResponseNewParams{
Model: "gpt-6.2-astra",
Background: openai.Bool(true),
Input: responses.ResponseNewParamsInputUnion{
OfString: openai.String("Write a very long novel about otters in space."),
},
})
if err != nil {
panic(err)
}
for response.Status == "queued" || response.Status == "in_progress" {
fmt.Println("Current status:", response.Status)
time.Sleep(2 * time.Second)
response, err = client.Responses.Get(context.Background(), response.ID, responses.ResponseGetParams{})
if err != nil {
panic(err)
}
}
fmt.Printf("Final status: %s\nOutput:\n%s\n", response.Status, response.OutputText())
}
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25import com.openai.client.OpenAIClient;
import com.openai.client.okhttp.OpenAIOkHttpClient;
import com.openai.models.responses.ResponseCreateParams;
import com.openai.models.responses.ResponseStatus;
ResponseCreateParams params =
ResponseCreateParams.builder()
.model("gpt-6.2-astra")
.input("Write a very long novel about otters in space.")
.background(true)
.build();
var response = client.responses().create(params);
while (response.status().filter(ResponseStatus.QUEUED::equals).isPresent()
|| response.status().filter(ResponseStatus.IN_PROGRESS::equals).isPresent()) {
System.out.println("Current status: " + response.status().orElseThrow());
Thread.sleep(1000);
response = client.responses().retrieve(response.id());
}
System.out.println("Final status: " + response.status().orElseThrow());
response.output().stream()
.flatMap(item -> item.message().stream())
.flatMap(message -> message.content().stream())
.flatMap(content -> content.outputText().stream())
.forEach(text -> System.out.println(text.text()));
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27using OpenAI.Responses;
#pragma warning disable OPENAI001
string key = Environment.GetEnvironmentVariable("OPENAI_API_KEY")!;
ResponsesClient client = new(key);
CreateResponseOptions options = new()
{
Model = "gpt-6.2-astra",
BackgroundModeEnabled = true,
};
options.InputItems.Add(
ResponseItem.CreateUserMessageItem("Write a very long novel about otters in space.")
);
ResponseResult response = await client.CreateResponseAsync(options);
while (response.Status is ResponseStatus.Queued or ResponseStatus.InProgress)
{
await Task.Delay(TimeSpan.FromSeconds(1));
response = await client.GetResponseAsync(response.Id);
}
if (response.Status != ResponseStatus.Completed)
{
throw new InvalidOperationException($"Background response ended with status: {response.Status}");
}
Console.WriteLine($"Status: {response.Status}");
Console.WriteLine(response.GetOutputText());
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18require "openai"
client = OpenAI::Client.new
response = client.responses.create(
model: "gpt-6.2-astra",
input: "Write a very long novel about otters in space.",
background: true
)
while [:queued, :in_progress].include?(response.status)
puts("Current status: #{response.status}")
sleep(2)
response = client.responses.retrieve(response.id)
end
puts("Final status: #{response.status}")
puts("Output:")
puts(response.output_text)
您也可以按以下方式取消正在生成的响应。
1
2
3curl -X POST https://api.openai.com/v1/responses/resp_123/cancel \
-H "Content-Type: application/json" \
-H "Authorization: Bearer $OPENAI_API_KEY"
1
2
3
4
5
6import OpenAI from "openai";
const client = new OpenAI();
const resp = await client.responses.cancel("resp_123");
console.log(resp.status);
1
2
3
4
5
6
7
8
9
10import os
from openai import OpenAI
response_id = os.environ["OPENAI_RESPONSE_ID"]
client = OpenAI()
resp = client.responses.cancel(response_id)
print(resp.status)
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19package main
import (
"context"
"fmt"
"github.com/openai/openai-go/v3"
)
func main() {
client := openai.NewClient()
canceled, err := client.Responses.Cancel(context.Background(), "resp_123")
if err != nil {
panic(err)
}
fmt.Println(canceled.Status)
}
1
2
3
4
5
6
7
8import com.openai.client.OpenAIClient;
import com.openai.client.okhttp.OpenAIOkHttpClient;
String responseId = "resp_123";
var response = client.responses().cancel(responseId);
System.out.println(response.status());
1
2
3
4
5
6
7
8
9
10
11using OpenAI.Responses;
#pragma warning disable OPENAI001
string key = Environment.GetEnvironmentVariable("OPENAI_API_KEY")!;
ResponsesClient client = new(key);
// Replace this illustrative ID with the background response to cancel.
string responseId = "resp_123";
ResponseResult response = await client.CancelResponseAsync(responseId);
Console.WriteLine(response.Status);
1
2
3
4
5require "openai"
client = OpenAI::Client.new
response = client.responses.cancel("resp_123")
puts(response.status)
取消操作具有幂等性。后续调用会返回最终的 Response 对象。要取消同步响应,请关闭连接。
要流式接收后台响应,请使用默认完成时限(soonest),并在创建响应时指定 stream: true。对于 2h 和 8h 请求,请省略 stream 或将其设为 false。
如果连接断开,请将最后收到的事件的 sequence_number 作为 starting_after 传入,以恢复事件流。只有在创建响应时启用了流式传输,之后才能流式接收该响应。
后台响应的首个 Token 等待时间可能比同步响应更长。
1
2
3
4
5
6
7
8
9
10
11
12
13
14curl https://api.openai.com/v1/responses \
-H "Content-Type: application/json" \
-H "Authorization: Bearer $OPENAI_API_KEY" \
-d '{
"model": "gpt-6.2-astra",
"input": "Write a very long novel about otters in space.",
"background": true,
"stream": true
}'
# To resume:
curl "https://api.openai.com/v1/responses/resp_123?stream=true&starting_after=42" \
-H "Content-Type: application/json" \
-H "Authorization: Bearer $OPENAI_API_KEY"
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25import OpenAI from "openai";
const client = new OpenAI();
const stream = await client.responses.create({
model: "gpt-6.2-astra",
input: "Write a very long novel about otters in space.",
background: true,
stream: true,
});
let cursor = null;
let responseId = null;
for await (const event of stream) {
console.log(event);
// Keepalives use connection counters, not stored-event resume positions.
if (String(event.type) !== "keepalive") cursor = event.sequence_number;
if ("response" in event && event.response?.id) responseId = event.response.id;
}
// If the connection drops, resume streaming from the last stored-event cursor:
// const resumedStream = await client.responses.retrieve(responseId, {
// stream: true,
// starting_after: cursor,
// });
// for await (const event of resumedStream) { ... }
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28from openai import OpenAI
client = OpenAI()
# Fire off an async response but also start streaming immediately
stream = client.responses.create(
model="gpt-6.2-astra",
input="Write a very long novel about otters in space.",
background=True,
stream=True,
)
cursor = None
response_id = None
for event in stream:
print(event)
if event.type == "response.created":
response_id = event.response.id
# Keepalives use connection counters, not stored-event resume positions.
if event.type != "keepalive":
cursor = event.sequence_number
# If your connection drops, the response continues running and you can reconnect:
# with client.responses.stream(response_id=response_id, starting_after=cursor) as resumed:
# for event in resumed:
# print(event)
if event.type == "response.created":
response_id = event.response.id
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49package main
import (
"context"
"fmt"
"github.com/openai/openai-go/v3"
"github.com/openai/openai-go/v3/responses"
)
func main() {
client := openai.NewClient()
stream := client.Responses.NewStreaming(context.Background(), responses.ResponseNewParams{
Model: "gpt-6.2-astra",
Background: openai.Bool(true),
Input: responses.ResponseNewParamsInputUnion{
OfString: openai.String("Write a very long novel about otters in space."),
},
})
defer stream.Close()
var cursor int64
var responseID string
for stream.Next() {
event := stream.Current()
fmt.Println(event.Type)
// Keepalives use connection counters, not stored-event resume positions.
if event.Type != "keepalive" {
cursor = event.SequenceNumber
}
if event.Response.ID != "" {
responseID = event.Response.ID
}
}
if err := stream.Err(); err != nil {
panic(err)
}
fmt.Printf("response %s last cursor %d\n", responseID, cursor)
// If the connection drops, resume streaming from the last cursor:
// resumed := client.Responses.GetStreaming(
// context.Background(),
// responseID,
// responses.ResponseGetParams{StartingAfter: openai.Int(cursor)},
// )
// for resumed.Next() {
// fmt.Println(resumed.Current().Type)
// }
}
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106import com.fasterxml.jackson.databind.json.JsonMapper;
import com.openai.client.OpenAIClient;
import com.openai.client.okhttp.OpenAIOkHttpClient;
import com.openai.core.RequestOptions;
import com.openai.core.http.StreamResponse;
import com.openai.errors.OpenAIIoException;
import com.openai.models.responses.ResponseCreateParams;
import com.openai.models.responses.ResponseRetrieveParams;
import com.openai.models.responses.ResponseStreamEvent;
import java.time.Duration;
import java.util.concurrent.atomic.AtomicBoolean;
import java.util.concurrent.atomic.AtomicLong;
import java.util.concurrent.atomic.AtomicReference;
ResponseCreateParams params =
ResponseCreateParams.builder()
.model("gpt-6.2-astra")
.input("Write a very long novel about otters in space.")
.background(true)
.build();
AtomicLong lastSequenceNumber = new AtomicLong(-1);
AtomicReference<String> responseId = new AtomicReference<>("");
AtomicBoolean streamCompleted = new AtomicBoolean(false);
JsonMapper json = new JsonMapper();
// This long-running response can exceed the SDK's default request deadline.
RequestOptions streamOptions = RequestOptions.builder().timeout(Duration.ofMinutes(30)).build();
try (StreamResponse<ResponseStreamEvent> stream =
client.responses().createStreaming(params, streamOptions)) {
stream.stream()
.forEach(
event -> {
if (event.failed().isPresent()
|| event.incomplete().isPresent()
|| event.error().isPresent()) {
throw new IllegalStateException("Background response did not complete");
}
var data = json.valueToTree(event);
// Keepalives count connection traffic, not stored response events.
if (data.path("type").asText().startsWith("response.")
&& data.path("sequence_number").isIntegralNumber()) {
lastSequenceNumber.set(data.path("sequence_number").asLong());
}
event
.created()
.ifPresent(
created -> {
responseId.set(created.response().id());
System.out.println("response.created");
});
event
.outputTextDelta()
.ifPresent(
delta -> {
System.out.println("response.output_text.delta");
});
event
.completed()
.ifPresent(
completed -> {
streamCompleted.set(true);
System.out.println("response.completed");
});
});
} catch (OpenAIIoException error) {
// Resume an interrupted stream once, from its stored-event checkpoint.
if (Thread.currentThread().isInterrupted()
|| streamCompleted.get()
|| responseId.get().isBlank()
|| lastSequenceNumber.get() < 0) {
throw error;
}
}
System.out.println(
"Response " + responseId.get() + "; last sequence number " + lastSequenceNumber.get());
if (!streamCompleted.get()) {
if (Thread.currentThread().isInterrupted()
|| responseId.get().isBlank()
|| lastSequenceNumber.get() < 0) {
throw new IllegalStateException("Stream ended without a resumable checkpoint");
}
try (StreamResponse<ResponseStreamEvent> resumed =
client
.responses()
.retrieveStreaming(
ResponseRetrieveParams.builder()
.responseId(responseId.get())
.startingAfter(lastSequenceNumber.get())
.build(),
streamOptions)) {
resumed.stream()
.forEach(
event -> {
if (event.failed().isPresent()
|| event.incomplete().isPresent()
|| event.error().isPresent()) {
throw new IllegalStateException("Resumed response did not complete");
}
event.outputTextDelta().ifPresent(delta -> System.out.println(delta.delta()));
event.completed().ifPresent(completed -> streamCompleted.set(true));
});
}
if (!streamCompleted.get()) {
throw new IllegalStateException("Resumed response did not complete");
}
}
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82using OpenAI.Responses;
#pragma warning disable OPENAI001
string key = Environment.GetEnvironmentVariable("OPENAI_API_KEY")!;
ResponsesClient client = new(key);
CreateResponseOptions options = new()
{
Model = "gpt-6.2-astra",
BackgroundModeEnabled = true,
StreamingEnabled = true,
};
options.InputItems.Add(
ResponseItem.CreateUserMessageItem("Write a very long novel about otters in space.")
);
string? responseId = null;
int lastSequenceNumber = -1;
bool completed = false;
void HandleUpdate(StreamingResponseUpdate update)
{
lastSequenceNumber = update.SequenceNumber;
switch (update)
{
case StreamingResponseCreatedUpdate created:
responseId = created.Response.Id;
break;
case StreamingResponseOutputTextDeltaUpdate text:
Console.Write(text.Delta);
break;
case StreamingResponseCompletedUpdate:
completed = true;
break;
case StreamingResponseFailedUpdate:
throw new InvalidOperationException("The background response failed.");
case StreamingResponseIncompleteUpdate:
throw new InvalidOperationException("The background response was incomplete.");
case StreamingResponseErrorUpdate error:
throw new InvalidOperationException($"The response stream failed: {error.Message}");
}
}
try
{
await foreach (
StreamingResponseUpdate update in client.CreateResponseStreamingAsync(options)
)
{
HandleUpdate(update);
}
}
catch (Exception error)
when (error is HttpRequestException or IOException && responseId is not null)
{
// The background response continues after its streaming connection is interrupted.
}
if (!completed)
{
if (responseId is null)
{
throw new InvalidOperationException("The response stream ended before providing its ID.");
}
GetResponseOptions resumeOptions = new(responseId)
{
StartingAfter = lastSequenceNumber,
StreamingEnabled = true,
};
await foreach (StreamingResponseUpdate update in client.GetResponseStreamingAsync(resumeOptions))
{
HandleUpdate(update);
}
if (!completed)
{
throw new InvalidOperationException(
"The resumed response stream ended before the background response completed."
);
}
}
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31require "openai"
# An uncapped background novel can run beyond the SDK's default ten minutes.
client = OpenAI::Client.new(timeout: 30 * 60)
stream = client.responses.stream(
model: "gpt-6.2-astra",
input: "Write a very long novel about otters in space.",
background: true
)
last_sequence_number = -1
response_id = ""
begin
stream.each do |event|
puts(event.type)
# The SDK filters transport keepalives before yielding stored events.
last_sequence_number = event.sequence_number || last_sequence_number
if event.is_a?(OpenAI::Models::Responses::ResponseCreatedEvent)
response_id = event.response.id
end
end
ensure
stream.close
end
puts("response #{response_id} last cursor #{last_sequence_number}")
# If the connection drops, resume from the last sequence number:
# client.responses.stream(response_id: response_id, starting_after: last_sequence_number).each do |event|
# puts(event.type)
# end