在長時間的互動中,你可以透過壓縮縮小上下文,同時保留後續回合所需的狀態。
隨著對話變長,壓縮有助於兼顧品質、成本與延遲。若要瞭解如何在變更對話紀錄時保留推理,請參閱保持前綴不變。
你可以在建立回應的請求(POST /responses 或 client.responses.create)中設定 context_management,並在其中指定 compact_threshold,以啟用伺服器端壓縮。
- 當上下文渲染後的 Token 數超過設定的門檻時,伺服器就會執行壓縮。
- 在此模式下,無須另外呼叫
/responses/compact。
- 回應串流會包含加密的壓縮項目。
- 關於 ZDR:在建立回應的請求中設定
store=false 後,即可搭配零資料保留(ZDR)使用伺服器端壓縮。
傳回的壓縮項目會以較少的 Token 將先前的關鍵狀態與推理保留到下一次執行。其內容不透明,並非供人直接解讀。
若以無狀態方式串接輸入陣列,請照常將輸出項目附加到陣列中。若使用 previous_response_id,每回合只需傳入新的使用者訊息。這兩種方式都會透過壓縮項目保留下一個視窗所需的上下文。
降低延遲的做法:將輸出項目附加到先前的輸入項目後,你可以移除最近一次壓縮項目之前的所有項目,以縮小請求並降低長尾延遲。最新的壓縮項目已保留繼續對話所需的上下文。若使用 previous_response_id 串接回應,請勿手動刪減項目。
- 照常呼叫
/responses,並在請求中加入 context_management,於其中指定 compact_threshold 以啟用伺服器端壓縮。
- 在串流傳送回應期間,若上下文大小超過門檻,伺服器就會執行一次壓縮,在同一串流中送出壓縮輸出項目,並在刪減上下文後繼續推論。
- 選擇一種方式繼續執行迴圈。若以無狀態方式串接輸入陣列,請將輸出項目(含壓縮項目)附加到下一個輸入陣列。若以
previous_response_id 串接回應,每回合只需傳入新的使用者訊息,並沿用上一個回應的 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
40import OpenAI from "openai";
import { toResponseInputItems } from "openai/lib/responses/ResponseInputItems";
import { createInterface } from "node:readline";
const client = new OpenAI();
const conversation: OpenAI.Responses.ResponseInput = [
{
type: "message",
role: "user",
content: "Let's begin a long coding task.",
},
];
// Subscribe before generation so piped follow-up turns are not lost.
const lines = createInterface({ input: process.stdin, crlfDelay: Infinity });
const turns = lines[Symbol.asyncIterator]();
async function respond() {
const response = await client.responses.create({
model: "gpt-5.3-codex",
input: conversation,
store: false,
context_management: [{ type: "compaction", compact_threshold: 200_000 }],
});
conversation.push(...toResponseInputItems(response.output));
console.log(response.output_text);
}
try {
await respond();
for await (const line of turns) {
if (!line.trim()) continue;
conversation.push({ type: "message", role: "user", content: line });
await respond();
}
} finally {
lines.close();
}
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
29conversation = [
{
"type": "message",
"role": "user",
"content": "Let's begin a long coding task.",
}
]
while keep_going:
response = client.responses.create(
model="gpt-5.3-codex",
input=conversation,
store=False,
context_management=[{"type": "compaction", "compact_threshold": 200000}],
)
conversation.extend(response.output)
try:
next_input = get_next_user_input()
except EOFError:
break
conversation.append(
{
"type": "message",
"role": "user",
"content": next_input,
}
)
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
56package main
import (
"bufio"
"context"
"encoding/json"
"fmt"
"os"
"github.com/openai/openai-go/v3"
"github.com/openai/openai-go/v3/responses"
)
func main() {
client := openai.NewClient()
conversation := []responses.ResponseInputItemUnionParam{
responses.ResponseInputItemParamOfMessage("Let's begin a long coding task.", responses.EasyInputMessageRoleUser),
}
scanner := bufio.NewScanner(os.Stdin)
for {
response, err := client.Responses.New(context.Background(), responses.ResponseNewParams{
Model: "gpt-5.3-codex",
Store: openai.Bool(false),
Input: responses.ResponseNewParamsInputUnion{OfInputItemList: conversation},
ContextManagement: []responses.ResponseNewParamsContextManagement{{
Type: "compaction", CompactThreshold: openai.Int(200000),
}},
})
if err != nil {
panic(err)
}
conversation = append(conversation, outputAsInput(response.Output)...)
fmt.Println(response.OutputText())
if !scanner.Scan() {
break
}
conversation = append(conversation,
responses.ResponseInputItemParamOfMessage(scanner.Text(), responses.EasyInputMessageRoleUser),
)
}
if err := scanner.Err(); err != nil {
panic(err)
}
}
func outputAsInput(output []responses.ResponseOutputItemUnion) []responses.ResponseInputItemUnionParam {
input := make([]responses.ResponseInputItemUnionParam, 0, len(output))
for _, item := range output {
var converted responses.ResponseInputItemUnion
if err := json.Unmarshal([]byte(item.RawJSON()), &converted); err != nil {
panic(err)
}
input = append(input, converted.ToParam())
}
return input
}
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
48import com.openai.client.OpenAIClient;
import com.openai.client.okhttp.OpenAIOkHttpClient;
import com.openai.core.JsonValue;
import com.openai.models.responses.EasyInputMessage;
import com.openai.models.responses.ResponseCreateParams;
import com.openai.models.responses.ResponseInputItem;
import java.util.ArrayList;
import java.util.List;
import java.util.Map;
var conversation = new ArrayList<ResponseInputItem>();
conversation.add(
ResponseInputItem.ofEasyInputMessage(
EasyInputMessage.builder()
.role(EasyInputMessage.Role.USER)
.content("Let's begin a long coding task.")
.build()));
ResponseCreateParams params =
ResponseCreateParams.builder()
.model("gpt-5.3-codex")
.inputOfResponse(conversation)
.store(false)
.putAdditionalBodyProperty(
"context_management",
JsonValue.from(List.of(Map.of("type", "compaction", "compact_threshold", 200000))))
.build();
var response = client.responses().create(params);
response.output().stream()
.map(item -> JsonValue.from(item).convert(ResponseInputItem.class))
.forEach(conversation::add);
conversation.add(
ResponseInputItem.ofEasyInputMessage(
EasyInputMessage.builder()
.role(EasyInputMessage.Role.USER)
.content("Now implement the next step.")
.build()));
client
.responses()
.create(params.toBuilder().inputOfResponse(conversation).build())
.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
27
28
29
30
31
32
33
34require "openai"
client = OpenAI::Client.new
conversation = [
{
type: :message,
role: :user,
content: "Let's begin a long coding task."
}
]
loop do
response = client.responses.create(
model: "gpt-5.3-codex",
input: conversation,
store: false,
context_management: [
{
type: :compaction,
compact_threshold: 200_000
}
]
)
conversation.concat(response.output)
puts(response.output_text)
line = $stdin.gets
break unless line
conversation << {
type: :message,
role: :user,
content: line.chomp
}
end
若要自行控制壓縮時機,可以在長時間執行的工作流程中使用獨立壓縮端點執行無狀態壓縮。
此端點完全無狀態,可搭配 ZDR 使用。
傳送完整的上下文視窗(包含訊息、工具及其他項目)後,端點會傳回新的壓縮後上下文視窗,供你傳入下一次 /responses 呼叫。
傳回的視窗包含一個加密的壓縮項目,以較少的 Token 保留先前的關鍵狀態與推理。該項目的內容不透明,並非供人直接解讀。
請注意,壓縮後的視窗通常不只包含壓縮項目,也可能包含從上一個視窗保留下來的項目。
處理輸出時,請勿刪減 /responses/compact 的輸出。傳回的視窗就是下一次呼叫應使用的上下文視窗,請將其原樣傳入下一次 /responses 呼叫。
- 照常呼叫
/responses,傳送包含使用者訊息、助理輸出及工具互動的輸入項目。
- 當上下文視窗變大時,呼叫
/responses/compact 以產生新的壓縮後上下文視窗。傳送至 /responses/compact 的視窗仍須在模型的上下文視窗容量範圍內。
- 後續呼叫
/responses 時,請以傳回的壓縮後視窗(含壓縮項目)取代完整對話紀錄,作為輸入。
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
26import OpenAI from "openai";
import { toResponseInputItems } from "openai/lib/responses/ResponseInputItems";
const client = new OpenAI();
const conversation: OpenAI.Responses.ResponseInput = [
{ role: "user", content: "Plan a trip to Kyoto." },
];
const compacted = await client.responses.compact({
model: "gpt-6-astra",
input: conversation,
});
const nextInput: OpenAI.Responses.ResponseInput = [
...toResponseInputItems(compacted.output),
{ role: "user", content: "Add two more days to the itinerary." },
];
const response = await client.responses.create({
model: "gpt-6-astra",
input: nextInput,
store: false,
});
console.log(response.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# Full window collected from prior turns
long_input_items_array = [{"role": "user", "content": "Plan a trip to Kyoto."}]
# 1) Compact the current window
compacted = client.responses.compact(
model="gpt-6-astra",
input=long_input_items_array,
)
# 2) Start the next turn by appending a new user message
next_input = [
*compacted.output, # Use compact output as-is
{
"type": "message",
"role": "user",
"content": user_input_message(),
},
]
next_response = client.responses.create(
model="gpt-6-astra",
input=next_input,
store=False, # Keep the flow ZDR-friendly
)
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
54package main
import (
"bufio"
"context"
"encoding/json"
"fmt"
"os"
"github.com/openai/openai-go/v3"
"github.com/openai/openai-go/v3/responses"
)
func main() {
client := openai.NewClient()
longInputItems := []responses.ResponseInputItemUnionParam{
responses.ResponseInputItemParamOfMessage("Plan a trip to Kyoto.", responses.EasyInputMessageRoleUser),
}
compacted, err := client.Responses.Compact(context.Background(), responses.ResponseCompactParams{
Model: "gpt-6-astra",
Input: responses.ResponseCompactParamsInputUnion{OfResponseInputItemArray: longInputItems},
})
if err != nil {
panic(err)
}
scanner := bufio.NewScanner(os.Stdin)
if !scanner.Scan() {
return
}
nextInput := append(outputAsInput(compacted.Output),
responses.ResponseInputItemParamOfMessage(scanner.Text(), responses.EasyInputMessageRoleUser),
)
nextResponse, err := client.Responses.New(context.Background(), responses.ResponseNewParams{
Model: "gpt-6-astra",
Store: openai.Bool(false),
Input: responses.ResponseNewParamsInputUnion{OfInputItemList: nextInput},
})
if err != nil {
panic(err)
}
fmt.Println(nextResponse.OutputText())
}
func outputAsInput(output []responses.ResponseOutputItemUnion) []responses.ResponseInputItemUnionParam {
input := make([]responses.ResponseInputItemUnionParam, 0, len(output))
for _, item := range output {
var converted responses.ResponseInputItemUnion
if err := json.Unmarshal([]byte(item.RawJSON()), &converted); err != nil {
panic(err)
}
input = append(input, converted.ToParam())
}
return input
}
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
52import com.openai.client.OpenAIClient;
import com.openai.client.okhttp.OpenAIOkHttpClient;
import com.openai.models.responses.EasyInputMessage;
import com.openai.models.responses.ResponseCompactParams;
import com.openai.models.responses.ResponseCompactionItemParam;
import com.openai.models.responses.ResponseCreateParams;
import com.openai.models.responses.ResponseInputItem;
import java.util.ArrayList;
var compacted =
client
.responses()
.compact(
ResponseCompactParams.builder()
.model("gpt-6-astra")
.input("Plan a trip to Kyoto.")
.build());
var input = new ArrayList<ResponseInputItem>();
for (var item : compacted.output()) {
item.message().map(ResponseInputItem::ofResponseOutputMessage).ifPresent(input::add);
item.reasoning().map(ResponseInputItem::ofReasoning).ifPresent(input::add);
item.compaction()
.map(
value ->
ResponseInputItem.ofCompaction(
ResponseCompactionItemParam.builder()
.id(value.id())
.encryptedContent(value.encryptedContent())
.build()))
.ifPresent(input::add);
}
input.add(
ResponseInputItem.ofEasyInputMessage(
EasyInputMessage.builder()
.role(EasyInputMessage.Role.USER)
.content("Add restaurant recommendations.")
.build()));
client
.responses()
.create(
ResponseCreateParams.builder()
.model("gpt-6-astra")
.inputOfResponse(input)
.store(false)
.build())
.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
27require "openai"
client = OpenAI::Client.new
long_input = [
{
role: :user,
content: "Plan a trip to Kyoto."
}
]
compaction = client.responses.compact(
model: "gpt-6-astra",
input: long_input
)
next_input = [
*compaction.output,
{
type: :message,
role: :user,
content: "Add restaurant recommendations."
}
]
response = client.responses.create(
model: "gpt-6-astra",
input: next_input,
store: false
)
puts(response.output_text)