规划研究任务

post/knowledge-fs/spaces/{knowledge_base_id}/research-tasks/plan

Request example

curl --request POST \
  --url https://api.dify.ai/v1/knowledge-fs/spaces/{knowledge_base_id}/research-tasks/plan \
  --header 'Authorization: Bearer <token>' \
  --header 'Content-Type: application/json' \
  --data '{
  "budgetUsd": 0.05,
  "mode": "research",
  "query": "What changed in the 2025 refund policy, and which regions are affected?",
  "topK": 10
}'

import requests

url = "https://api.dify.ai/v1/knowledge-fs/spaces/{knowledge_base_id}/research-tasks/plan"
headers = {
    "Authorization": "Bearer <token>",
    "Content-Type": "application/json",
}
payload = {
    "budgetUsd": 0.05,
    "mode": "research",
    "query": "What changed in the 2025 refund policy, and which regions are affected?",
    "topK": 10
}

response = requests.request("POST", url, headers=headers, json=payload)

print(response.text)

const response = await fetch('https://api.dify.ai/v1/knowledge-fs/spaces/{knowledge_base_id}/research-tasks/plan', {
  method: 'POST',
  headers: {
    'Authorization': 'Bearer <token>',
    'Content-Type': 'application/json'
  },
  body: JSON.stringify({
    "budgetUsd": 0.05,
    "mode": "research",
    "query": "What changed in the 2025 refund policy, and which regions are affected?",
    "topK": 10
  })
});

const data = await response.json();
console.log(data);

<?php

$curl = curl_init();

curl_setopt_array($curl, [
  CURLOPT_URL => 'https://api.dify.ai/v1/knowledge-fs/spaces/{knowledge_base_id}/research-tasks/plan',
  CURLOPT_RETURNTRANSFER => true,
  CURLOPT_CUSTOMREQUEST => 'POST',
  CURLOPT_POSTFIELDS => '{
  "budgetUsd": 0.05,
  "mode": "research",
  "query": "What changed in the 2025 refund policy, and which regions are affected?",
  "topK": 10
}',
  CURLOPT_HTTPHEADER => [
    'Authorization: Bearer <token>',
    'Content-Type: application/json',
  ],
]);

$response = curl_exec($curl);
curl_close($curl);

echo $response;

package main

import (
	"fmt"
	"io"
	"net/http"
	"strings"
)

func main() {
	url := "https://api.dify.ai/v1/knowledge-fs/spaces/{knowledge_base_id}/research-tasks/plan"
	payload := strings.NewReader("{\n  \"budgetUsd\": 0.05,\n  \"mode\": \"research\",\n  \"query\": \"What changed in the 2025 refund policy, and which regions are affected?\",\n  \"topK\": 10\n}")

	req, _ := http.NewRequest("POST", url, payload)
	req.Header.Add("Authorization", "Bearer <token>")
	req.Header.Add("Content-Type", "application/json")

	res, _ := http.DefaultClient.Do(req)
	defer res.Body.Close()
	body, _ := io.ReadAll(res.Body)

	fmt.Println(string(body))
}

HttpRequest request = HttpRequest.newBuilder()
    .uri(URI.create("https://api.dify.ai/v1/knowledge-fs/spaces/{knowledge_base_id}/research-tasks/plan"))
    .header("Authorization", "Bearer <token>")
    .header("Content-Type", "application/json")
    .method("POST", HttpRequest.BodyPublishers.ofString("{\n  \"budgetUsd\": 0.05,\n  \"mode\": \"research\",\n  \"query\": \"What changed in the 2025 refund policy, and which regions are affected?\",\n  \"topK\": 10\n}"))
    .build();

HttpResponse<String> response = HttpClient.newHttpClient()
    .send(request, HttpResponse.BodyHandlers.ofString());

System.out.println(response.body());

require 'uri'
require 'net/http'

url = URI('https://api.dify.ai/v1/knowledge-fs/spaces/{knowledge_base_id}/research-tasks/plan')

http = Net::HTTP.new(url.host, url.port)
http.use_ssl = url.scheme == 'https'

request = Net::HTTP::Post.new(url)
request['Authorization'] = 'Bearer <token>'
request['Content-Type'] = 'application/json'
request.body = '{
  "budgetUsd": 0.05,
  "mode": "research",
  "query": "What changed in the 2025 refund policy, and which regions are affected?",
  "topK": 10
}'

response = http.request(request)
puts response.read_body

Response

{
  "budget": {
    "budget_usd": 0.05,
    "exceeds_budget": false,
    "remaining_budget_usd": 0.018238
  },
  "estimates": {
    "cacheHitProbability": 0.25,
    "costUsd": {
      "currency": "USD",
      "estimated": 0.031762,
      "max": 0.042879,
      "min": 0.020645
    },
    "inputTokens": 3070,
    "latencyMs": {
      "p50": 2800,
      "p95": 5040
    },
    "outputTokens": 1776,
    "retrievalSteps": 3,
    "scannedResources": 620,
    "toolCalls": 6,
    "totalTokens": 4846,
    "workBounds": {
      "modelCalls": {
        "estimated": 3,
        "max": 4,
        "min": 2
      },
      "openedResources": {
        "estimated": 20,
        "max": 40,
        "min": 0
      },
      "retrievalSteps": {
        "estimated": 3,
        "max": 5,
        "min": 1
      }
    }
  },
  "knowledge_space_id": "b2f1d3a4-5e6c-4a7b-8c9d-0e1f2a3b4c5d",
  "query": "What changed in the 2025 refund policy, and which regions are affected?",
  "retrieval_plan": {
    "dense_top_k": 100,
    "fts_top_k": 100,
    "fusion_limit": 50,
    "query_language": "latin",
    "requested_mode": "research",
    "rerank_candidate_limit": 50,
    "resolved_mode": "research",
    "strategy_version": "retrieval-planner-v2",
    "top_k": 10
  },
  "steps": [
    {
      "estimatedCostUsd": 0.003126,
      "estimatedInputTokens": 274,
      "estimatedLatencyMs": 350,
      "estimatedOutputTokens": 192,
      "estimatedToolCalls": 1,
      "name": "plan"
    },
    {
      "estimatedCostUsd": 0.00124,
      "estimatedInputTokens": 0,
      "estimatedLatencyMs": 600,
      "estimatedOutputTokens": 0,
      "estimatedToolCalls": 3,
      "name": "retrieve"
    },
    {
      "estimatedCostUsd": 0.007542,
      "estimatedInputTokens": 978,
      "estimatedLatencyMs": 650,
      "estimatedOutputTokens": 384,
      "estimatedToolCalls": 1,
      "name": "analyze"
    },
    {
      "estimatedCostUsd": 0.019854,
      "estimatedInputTokens": 1818,
      "estimatedLatencyMs": 1200,
      "estimatedOutputTokens": 1200,
      "estimatedToolCalls": 1,
      "name": "generate"
    }
  ],
  "strategy_version": "research-dry-run-planner-v1"
}

{
  "code": "knowledge_fs_invalid_request",
  "message": "Agent Knowledge Base request is invalid.",
  "status": 400,
  "violations": [
    {
      "field": [
        "limits"
      ],
      "type": "extra_forbidden"
    }
  ]
}

{
  "code": "knowledge_fs_invalid_request",
  "failure": {
    "category": "validation",
    "code": "KNOWLEDGE_FS_INVALID_REQUEST",
    "message": "The Agent Knowledge Base request is invalid.",
    "retryPolicy": "never",
    "traceId": "d44c6dd1-da9c-46e7-81e1-5f09707c4352"
  },
  "message": "The Agent Knowledge Base request is invalid.",
  "status": 400
}

{
  "code": "knowledge_fs_invalid_credential",
  "message": "Invalid Agent Knowledge Base service credential.",
  "status": 401
}

{
  "code": "unauthorized",
  "message": "Access token is invalid",
  "status": 401
}

{
  "code": "forbidden",
  "message": "Sorry, you have reached the knowledge base request rate limit of your subscription.",
  "status": 403
}

{
  "code": "knowledge_fs_access_denied",
  "message": "Agent Knowledge Base operation is not allowed.",
  "status": 403
}

{
  "code": "knowledge_fs_resource_not_found",
  "failure": {
    "category": "not_found",
    "code": "KNOWLEDGE_FS_NOT_FOUND",
    "message": "The requested Agent Knowledge Base resource was not found.",
    "retryPolicy": "never",
    "traceId": "8f3a2c14-9d6b-4c07-b5a1-2e7d0c93f5b2"
  },
  "message": "The requested Agent Knowledge Base resource was not found.",
  "status": 404
}

{
  "code": "not_found",
  "message": "The requested URL was not found on the server. If you entered the URL manually please check your spelling and try again.",
  "status": 404
}

{
  "code": "knowledge_fs_request_too_large",
  "message": "Agent Knowledge Base request is too large.",
  "status": 413
}

{
  "code": "knowledge_fs_upstream_unavailable",
  "failure": {
    "action": "contact_admin",
    "category": "internal",
    "code": "KNOWLEDGE_FS_INTERNAL_ERROR",
    "message": "Agent Knowledge Base could not complete the operation. Try again, or contact an administrator with the error reference.",
    "retryPolicy": "manual",
    "traceId": "5e2a9c17-3b4d-4f60-8a1b-2c3d4e5f6a7b"
  },
  "message": "Agent Knowledge Base could not complete the operation. Try again, or contact an administrator with the error reference.",
  "status": 502
}

{
  "code": "knowledge_fs_upstream_unavailable",
  "message": "Agent Knowledge Base is unavailable.",
  "status": 502
}

{
  "code": "knowledge_fs_operation_unavailable",
  "message": "Agent Knowledge Base operation is not available.",
  "status": 503
}

{
  "code": "knowledge_fs_operation_unavailable",
  "failure": {
    "action": "retry",
    "category": "dependency",
    "code": "KNOWLEDGE_FS_UNAVAILABLE",
    "message": "A service required by Agent Knowledge Base is temporarily unavailable.",
    "retryPolicy": "manual",
    "traceId": "5e2a9c17-3b4d-4f60-8a1b-2c3d4e5f6a7b"
  },
  "message": "A service required by Agent Knowledge Base is temporarily unavailable.",
  "status": 503
}

post/knowledge-fs/spaces/{knowledge_base_id}/research-tasks/plan

Authorization

Authorizationstringheaderrequired#

每个请求都通过 API Key 认证：Authorization: Bearer {API_KEY}。应用接口使用应用 API Key，知识库接口使用知识库 API Key（快速开始）。

API Key 应保存在服务端，切勿嵌入客户端代码。缺失或无效的 Key 会返回 HTTP 401（unauthorized）。

Path parameters

knowledge_base_idstringrequired#

知识库 ID：即知识库 API 访问 对话框中所示服务 API 端点的最后一段（…/knowledge-fs/spaces/<knowledge base ID>），也是地址栏中 /datasets/new/ 之后的那一段。

Body

application/json

budgetUsdnumber#

用于与成本估算比较的花费上限（USD）。决定响应中的 budget.exceeds_budget 和 budget.remaining_budget_usd。

Range: >= 0

modeenum<string>#

检索深度：fast、deep 或 research（参见检索深度）。省略时使用知识库的默认深度，即获取知识库设置中的 retrieval.default_mode。

Options:fastdeepresearchnull

querystringrequired#

要查询的问题。不能为空白，且不超过 16,384 个 UTF-8 字节。

Length: max 16000Pattern: \S

topKinteger#

检索结果合并并重排序后最多保留的分段数。显式传入的值必须在 1～50 之间。省略时使用知识库已发布的 Top K 设置，该设置最高可达 100。值越大，模型获得的上下文越多，干扰也越多；值越小，结果越集中，但可能漏掉相关分段。

Range: 1–50

Responses

application/json

规划结果及其估算。

budgetobject#

成本估算与请求中 budgetUsd 的比较结果。

Show child attributes

budget_usdnumber#

请求中的 budgetUsd。未发送时为 null。

exceeds_budgetboolean#

estimates.costUsd.estimated 高于 budgetUsd 时为 true。未设置预算时始终为 false。仅供参考：创建研究任务仍会受理该任务，并在下一次模型调用可能超出预算时取消它。

remaining_budget_usdnumber#

budgetUsd 减去估算成本。超出预算时为负数。未设置预算时为 null。

estimatesobject#

任务的工作量、成本和延迟估算。可与你打算发送给创建研究任务的 limits 对比。

Show child attributes

cacheHitProbabilitynumber#

预计由缓存提供的检索比例，取值范围为 0 到 1。

costUsdobject#

任务的估算成本。

Show child attributes

currencystring#

始终为 USD。

estimatednumber#

点估计值（USD）。

maxnumber#

上限（USD）。

minnumber#

下限（USD）。

inputTokensinteger#

所有步骤预计消耗的模型输入 token 数。

latencyMsobject#

预计的端到端耗时，单位为毫秒。可根据 p50 安排首次轮询获取研究任务的时间。

Show child attributes

p50integer#

中位数估算。

p95integer#

第 95 百分位估算。limits.timeoutMs 对照的就是这个值。

outputTokensinteger#

所有步骤预计产生的模型输出 token 数。

retrievalStepsinteger#

预计的检索轮数。对照 limits.maxRetrievalSteps 检查。

scannedResourcesinteger#

预计读取的文档和章节数。对照 limits.maxScannedResources 检查。

toolCallsinteger#

预计的工具调用次数。对照 limits.maxToolCalls 检查。

totalTokensinteger#

inputTokens 与 outputTokens 之和。

workBoundsobject#

点估计值背后的范围。创建研究任务检查的是点估计值，不检查这些范围。

Show child attributes

modelCallsobject#

模型调用次数的估计值、最小值和最大值。

Show child attributes

estimatedinteger#

点估计值。

maxinteger#

上限。

mininteger#

下限。

openedResourcesobject#

打开的文档和章节数的估计值、最小值和最大值。

Show child attributes

estimatedinteger#

点估计值。

maxinteger#

上限。

mininteger#

下限。

retrievalStepsobject#

检索轮数的估计值、最小值和最大值。

Show child attributes

estimatedinteger#

点估计值。

maxinteger#

上限。

mininteger#

下限。

knowledge_space_idstring#

知识库的内部 ID，并非路径中的知识库 ID。

querystring#

去除首尾空白后的问题。

retrieval_planobject#

检索将如何执行。

Show child attributes

dense_top_kinteger#

从向量检索中选取的候选数。

fts_top_kinteger#

从关键词检索中选取的候选数。

fusion_limitinteger#

合并向量检索和关键词检索结果后保留的候选数。

query_languagestring#

检测到的问题语言：cjk、latin、mixed-cjk-latin 或 other。

requested_modestring#

请求中的 mode；省略时为知识库的默认深度。取值为 fast、deep 或 research。

rerank_candidate_limitinteger#

传给重排序模型的候选数。

resolved_modestring#

将要运行的深度：fast、deep 或 research。调用创建研究任务时发送相同的 mode。如果那里使用了不同的深度，你的 limits 将对照另一组估算检查。

strategy_versionstring#

生成此规划所用检索策略的版本。

top_kinteger#

检索结果合并并重排序后最多保留的分段数。

stepsobject[]#

各步骤的估算，按执行顺序排列。

Show child attributes

estimatedCostUsdnumber#

该步骤的估算成本（USD）。

estimatedInputTokensinteger#

预计的模型输入 token 数。

estimatedLatencyMsinteger#

预计耗时，单位为毫秒。

estimatedOutputTokensinteger#

预计的模型输出 token 数。

estimatedToolCallsinteger#

预计的工具调用次数。

namestring#

步骤名称。plan、retrieve、analyze 和 generate 始终出现，并按此顺序排列。检索策略需要读取文档大纲时，inspect 紧跟在 plan 之后。

strategy_versionstring#

估算器的版本。

创建研究任务获取研究任务