Skip to content

RAGFlow is an agent-native database: a Retrieval-Augmented Generation (RAG) engine that grew into context infrastructure.

Revert "feat(agent): support document_ids in retrieval node and tool" (#19837)

Wang Qi501424f01f3a
diff --git a/agent/tools/retrieval.py b/agent/tools/retrieval.py
index 7997dc1a98..961aaae105 100644
--- a/agent/tools/retrieval.py
+++ b/agent/tools/retrieval.py
@@ -16,7 +16,6 @@
import asyncio
from functools import partial
import json
-import logging
import os
import re
from abc import ABC
@@ -86,7 +85,6 @@ class RetrievalParam(ToolParamBase):
self.cross_languages = []
self.toc_enhance = False
self.meta_data_filter = {}
- self.document_ids = ""
def check(self):
self.check_decimal_float(self.similarity_threshold, "[Retrieval] Similarity threshold")
@@ -94,10 +92,7 @@ class RetrievalParam(ToolParamBase):
self.check_positive_number(self.top_n, "[Retrieval] Top N")
def get_input_form(self) -> dict[str, dict]:
- return {
- "query": {"name": "Query", "type": "line"},
- "document_ids": {"name": "Document IDs", "type": "line"},
- }
+ return {"query": {"name": "Query", "type": "line"}}
class Retrieval(ToolBase, ABC):
@@ -108,30 +103,6 @@ class Retrieval(ToolBase, ABC):
"""Get dataset IDs with backward compatibility for kb_ids."""
return self._param.dataset_ids or getattr(self._param, "kb_ids", None) or []
- @staticmethod
- def _normalize_document_ids(value) -> list[str]:
- if value is None or value == "" or value == []:
- return []
- if isinstance(value, list):
- return [str(v) for v in value if v]
- if isinstance(value, str):
- value = value.strip()
- if not value:
- return []
- return [value]
- return []
-
- def _resolve_document_ids(self, kwargs_document_ids=None) -> list[str]:
- doc_ids = self._normalize_document_ids(kwargs_document_ids)
- if doc_ids:
- return doc_ids
- raw = getattr(self._param, "document_ids", None)
- if raw is None or raw == "" or raw == []:
- return []
- if isinstance(raw, str) and self._canvas.is_reff(raw):
- return self._normalize_document_ids(self._canvas.get_variable_value(raw))
- return self._normalize_document_ids(raw)
-
def _resolve_manual_filter(self, flt: dict) -> dict:
# Return a new dict instead of mutating `flt` in place. The caller
# passes filters straight out of self._param.meta_data_filter, so
@@ -153,7 +124,7 @@ class Retrieval(ToolBase, ABC):
resolved["value"] = self._replace_template_matches(pat, content, replace)
return resolved
- async def _retrieve_kb(self, query_text: str, document_ids=None):
+ async def _retrieve_kb(self, query_text: str):
kb_ids: list[str] = []
for id in self._dataset_ids:
if id.find("@") < 0:
@@ -193,13 +164,7 @@ class Retrieval(ToolBase, ABC):
vars = {k: o["value"] for k, o in vars.items()}
query = self.string_format(query_text, vars)
- doc_ids = self._resolve_document_ids(document_ids)
- if doc_ids:
- logging.info("Retrieval resolved document_ids=%s", doc_ids)
- kb_doc_ids = KnowledgebaseService.list_documents_by_ids(filtered_kb_ids)
- for doc_id in doc_ids:
- if doc_id not in kb_doc_ids:
- raise Exception(f"The datasets don't own the document {doc_id}")
+ doc_ids = []
if self._param.meta_data_filter != {}:
# Defer the (potentially expensive) metadata table load — manual
# filters served by ES push-down never need it. The loader is
@@ -345,13 +310,12 @@ class Retrieval(ToolBase, ABC):
self.set_output("formalized_content", self._param.empty_response)
return
- document_ids = self._resolve_document_ids(kwargs.get("document_ids"))
if hasattr(self._param, "retrieval_from") and self._param.retrieval_from == "dataset":
- return await self._retrieve_kb(kwargs["query"], document_ids)
+ return await self._retrieve_kb(kwargs["query"])
elif hasattr(self._param, "retrieval_from") and self._param.retrieval_from == "memory":
return await self._retrieve_memory(kwargs["query"])
elif self._dataset_ids:
- return await self._retrieve_kb(kwargs["query"], document_ids)
+ return await self._retrieve_kb(kwargs["query"])
elif hasattr(self._param, "memory_ids") and self._param.memory_ids:
return await self._retrieve_memory(kwargs["query"])
else:
diff --git a/internal/agent/component/universe_a_wrappers.go b/internal/agent/component/universe_a_wrappers.go
index 504dc56dfd..999916c3b1 100644
--- a/internal/agent/component/universe_a_wrappers.go
+++ b/internal/agent/component/universe_a_wrappers.go
@@ -68,7 +68,6 @@ type retrievalParams struct {
Query string
KbIDs []string
MemoryIDs []string
- DocumentIDs []string
UserID string
TopN int
TopK int
@@ -110,13 +109,6 @@ func parseRetrievalParams(params map[string]any) retrievalParams {
out.KbIDs = append(out.KbIDs, v...)
}
out.MemoryIDs = toStringSlice(params["memory_ids"])
- if v, ok := params["document_ids"].(string); ok {
- if strings.TrimSpace(v) != "" {
- out.DocumentIDs = []string{v}
- }
- } else {
- out.DocumentIDs = toStringSlice(params["document_ids"])
- }
if v, ok := params["user_id"].(string); ok {
out.UserID = v
}
@@ -178,11 +170,10 @@ func (c *retrievalComponent) Name() string { return "Retrieval" }
func (c *retrievalComponent) Inputs() map[string]string {
return map[string]string{
- "query": "Natural-language search query.",
- "dataset_ids": "Optional list of dataset IDs to restrict the search to (overrides node-level kb_ids).",
- "document_ids": "Optional list of document IDs to restrict retrieval scope.",
- "top_n": "Maximum chunks to return (default 8, overrides node-level top_n).",
- "use_kg": "GraphRAG toggle (returns ErrKGRetrievalServiceMissing until a kg adapter is registered).",
+ "query": "Natural-language search query.",
+ "dataset_ids": "Optional list of dataset IDs to restrict the search to (overrides node-level kb_ids).",
+ "top_n": "Maximum chunks to return (default 8, overrides node-level top_n).",
+ "use_kg": "GraphRAG toggle (returns ErrKGRetrievalServiceMissing until a kg adapter is registered).",
}
}
@@ -192,10 +183,6 @@ func (c *retrievalComponent) GetInputForm() map[string]any {
"name": "Query",
"type": "line",
},
- "document_ids": map[string]any{
- "name": "Document IDs",
- "type": "line",
- },
}
}
@@ -307,16 +294,6 @@ func (c *retrievalComponent) applyDefaults(inputs map[string]any) map[string]any
if _, ok := out["memory_ids"]; !ok && len(c.params.MemoryIDs) > 0 {
out["memory_ids"] = append([]string(nil), c.params.MemoryIDs...)
}
- if _, ok := out["document_ids"]; !ok && len(c.params.DocumentIDs) > 0 {
- out["document_ids"] = append([]string(nil), c.params.DocumentIDs...)
- }
- if s, ok := out["document_ids"].(string); ok {
- if strings.TrimSpace(s) == "" {
- delete(out, "document_ids")
- } else {
- out["document_ids"] = []string{s}
- }
- }
if _, ok := out["user_id"]; !ok && c.params.UserID != "" {
out["user_id"] = c.params.UserID
}
diff --git a/internal/agent/tool/registry.go b/internal/agent/tool/registry.go
index 76ee35c691..7781eadcaa 100644
--- a/internal/agent/tool/registry.go
+++ b/internal/agent/tool/registry.go
@@ -399,7 +399,7 @@ func buildRetrievalTool(params map[string]any) (einotool.BaseTool, error) {
"keywords_similarity_weight", "use_kg", "rerank_id", "empty_response",
"toc_enhance", "meta_data_filter", "retrieval_from", "memory_ids",
"kb_vars", "cross_languages", "function_name", "description", "meta",
- "inputs", "outputs", "user_id", "document_ids":
+ "inputs", "outputs", "user_id":
default:
return nil, fmt.Errorf("agent tool: retrieval tool does not accept node-level param %s", key)
}
@@ -423,15 +423,6 @@ func buildRetrievalTool(params map[string]any) (einotool.BaseTool, error) {
} else if ok {
defaults.MemoryIDs = ids
}
- if v, ok := params["document_ids"].(string); ok {
- if strings.TrimSpace(v) != "" {
- defaults.DocumentIDs = []string{v}
- }
- } else if ids, ok, err := stringSliceParam(params, "document_ids"); err != nil {
- return nil, fmt.Errorf("agent tool: retrieval config: %w", err)
- } else if ok {
- defaults.DocumentIDs = ids
- }
if v, ok := stringParam(params, "user_id"); ok {
defaults.UserID = v
}
diff --git a/internal/agent/tool/retrieval.go b/internal/agent/tool/retrieval.go
index db399e6a48..f4d3b2bb51 100644
--- a/internal/agent/tool/retrieval.go
+++ b/internal/agent/tool/retrieval.go
@@ -60,7 +60,6 @@ type retrievalArgs struct {
DatasetIDs []string `json:"dataset_ids,omitempty"`
KBIDs []string `json:"kb_ids,omitempty"`
MemoryIDs []string `json:"memory_ids,omitempty"`
- DocumentIDs []string `json:"document_ids,omitempty"`
UserID string `json:"user_id,omitempty"`
TopN int `json:"top_n,omitempty"`
RerankCandidatesCount int `json:"rerank_candidates_count,omitempty"`
@@ -132,11 +131,6 @@ func (r *RetrievalTool) Info(_ context.Context) (*schema.ToolInfo, error) {
Desc: "The keywords to search the dataset. The keywords should be the most important words/terms (including synonyms) from the original request.",
Required: true,
},
- "document_ids": {
- Type: schema.Array,
- Desc: "Optional list of document IDs to restrict retrieval scope.",
- Required: false,
- },
}),
}, nil
}
@@ -166,7 +160,6 @@ func (r *RetrievalTool) InvokableRun(ctx context.Context, argumentsInJSON string
common.Debug("agent retrieval tool: parsed arguments",
zap.String("query", args.Query),
zap.Strings("dataset_ids", args.DatasetIDs),
- zap.Strings("document_ids", args.DocumentIDs),
zap.Int("top_n", args.TopN),
zap.Int("top_k", args.TopK),
zap.Float64p("keywords_similarity_weight", args.KeywordsSimilarityWeight),
@@ -198,11 +191,6 @@ func (r *RetrievalTool) InvokableRun(ctx context.Context, argumentsInJSON string
return "", err
}
args.DatasetIDs = resolvedDatasetIDs
- resolvedDocumentIDs, err := resolveRetrievalDocumentIDs(ctx, args.DocumentIDs)
- if err != nil {
- return "", err
- }
- args.DocumentIDs = resolvedDocumentIDs
resolvedFilter, err := resolveRetrievalFilter(ctx, args.MetaDataFilter)
if err != nil {
return "", err
@@ -229,7 +217,6 @@ func (r *RetrievalTool) InvokableRun(ctx context.Context, argumentsInJSON string
TOCEnhance: args.TOCEnhance,
MetaDataFilter: cloneStringAnyMap(args.MetaDataFilter),
RetrievalFrom: args.RetrievalFrom,
- DocScope: append([]string(nil), args.DocumentIDs...),
UserID: args.UserID,
TenantID: retrievalTenantID(ctx),
}
@@ -293,9 +280,6 @@ func (r *RetrievalTool) mergeDefaults(args retrievalArgs) retrievalArgs {
if len(args.MemoryIDs) == 0 && len(r.defaults.MemoryIDs) != 0 {
args.MemoryIDs = append([]string(nil), r.defaults.MemoryIDs...)
}
- if len(args.DocumentIDs) == 0 && len(r.defaults.DocumentIDs) != 0 {
- args.DocumentIDs = append([]string(nil), r.defaults.DocumentIDs...)
- }
if args.TopN <= 0 {
args.TopN = r.defaults.TopN
}
@@ -449,61 +433,6 @@ func resolveRetrievalDatasetIDs(ctx context.Context, datasetIDs []string) ([]str
return compactStrings(resolved), nil
}
-func resolveRetrievalDocumentIDs(ctx context.Context, documentIDs []string) ([]string, error) {
- state, _, err := runtime.GetStateFromContext[*runtime.CanvasState](ctx)
- if err != nil || state == nil {
- return compactStrings(documentIDs), nil
- }
- resolved := make([]string, 0, len(documentIDs))
- for _, documentID := range documentIDs {
- documentID = strings.TrimSpace(documentID)
- if documentID == "" {
- continue
- }
- if strings.Contains(documentID, "@") {
- value, getErr := state.GetVar(documentID)
- if getErr != nil {
- return nil, fmt.Errorf("retrieval: resolve document variable %q: %w", documentID, getErr)
- }
- resolved = append(resolved, normalizeResolvedDocumentIDs(value)...)
- continue
- }
- text, resolveErr := runtime.ResolveTemplateAuto(documentID, state)
- if resolveErr != nil {
- return nil, fmt.Errorf("retrieval: resolve document_ids template: %w", resolveErr)
- }
- resolved = append(resolved, normalizeResolvedDocumentIDs(text)...)
- }
- return compactStrings(resolved), nil
-}
-
-func normalizeResolvedDocumentIDs(value any) []string {
- switch typed := value.(type) {
- case nil:
- return nil
- case string:
- typed = strings.TrimSpace(typed)
- if typed == "" {
- return nil
- }
- return []string{typed}
- case []string:
- return typed
- case []any:
- out := make([]string, 0, len(typed))
- for _, item := range typed {
- out = append(out, normalizeResolvedDocumentIDs(item)...)
- }
- return out
- default:
- text := strings.TrimSpace(fmt.Sprint(typed))
- if text == "" || text == "<nil>" {
- return nil
- }
- return []string{text}
- }
-}
-
func resolveRetrievalFilter(ctx context.Context, filter map[string]any) (map[string]any, error) {
if filter == nil {
return nil, nil
diff --git a/internal/agent/tool/retrieval_service.go b/internal/agent/tool/retrieval_service.go
index 8459c6a5f9..e615056df5 100644
--- a/internal/agent/tool/retrieval_service.go
+++ b/internal/agent/tool/retrieval_service.go
@@ -84,8 +84,8 @@ type RetrievalRequest struct {
RankFeature *map[string]float64
MetaDataFilter map[string]any
RetrievalFrom string
- // DocScope restricts retrieval to a set of document ids (from document_ids
- // on the retrieval node/tool, or dataset_navigation_by_tree). Empty = no doc filter.
+ // DocScope restricts retrieval to a set of document ids (the doc_id list
+ // routed by the dataset_navigation_by_tree tool). Empty = no doc filter.
DocScope []string
// TenantID is the calling tenant (== user_id in RAGFlow's data model).
// It is used for dataset-name resolution and memory access. Reads from
diff --git a/test/unit_test/agent/test_retrieval_document_ids.py b/test/unit_test/agent/test_retrieval_document_ids.py
deleted file mode 100644
index 7e8964acea..0000000000
--- a/test/unit_test/agent/test_retrieval_document_ids.py
+++ /dev/null
@@ -1,33 +0,0 @@
-#
-# Copyright 2026 The InfiniFlow Authors. All Rights Reserved.
-#
-from types import SimpleNamespace
-
-from agent.tools.retrieval import Retrieval
-
-
-class TestNormalizeDocumentIds:
- def test_empty_values(self):
- assert Retrieval._normalize_document_ids(None) == []
- assert Retrieval._normalize_document_ids("") == []
- assert Retrieval._normalize_document_ids([]) == []
-
- def test_list(self):
- assert Retrieval._normalize_document_ids(["a", "b"]) == ["a", "b"]
- assert Retrieval._normalize_document_ids(["a", "", "b"]) == ["a", "b"]
-
- def test_single_string(self):
- assert Retrieval._normalize_document_ids("doc-1") == ["doc-1"]
-
-
-class TestResolveDocumentIds:
- def test_prefers_kwargs(self):
- tool = Retrieval.__new__(Retrieval)
- tool._param = SimpleNamespace(document_ids="doc-config")
- assert tool._resolve_document_ids(["doc-runtime"]) == ["doc-runtime"]
-
- def test_falls_back_to_param(self):
- tool = Retrieval.__new__(Retrieval)
- tool._param = SimpleNamespace(document_ids=["doc-config"])
- tool._canvas = SimpleNamespace(is_reff=lambda _exp: False)
- assert tool._resolve_document_ids(None) == ["doc-config"]
diff --git a/web/src/interfaces/database/agent.ts b/web/src/interfaces/database/agent.ts
index b114b0b193..9be9bc43fd 100644
--- a/web/src/interfaces/database/agent.ts
+++ b/web/src/interfaces/database/agent.ts
@@ -170,7 +170,6 @@ export interface IRetrievalForm {
rerank_id?: string;
tenant_rerank_id?: string;
dataset_ids?: string[];
- document_ids?: string | string[];
}
export interface ICodeForm {
diff --git a/web/src/locales/en.ts b/web/src/locales/en.ts
index 70b77337df..8abd489686 100644
--- a/web/src/locales/en.ts
+++ b/web/src/locales/en.ts
@@ -3367,9 +3367,6 @@ This process aggregates variables from multiple branches into a single variable
switchPromptMessage:
'The prompt words will change. Please confirm whether you want to discard the existing prompt words?',
queryRequired: 'Query is required',
- documentIds: 'Document IDs',
- documentIdsTip:
- 'Optional list of document IDs to restrict retrieval scope. Supports upstream variable references.',
queryTip: 'Select the variable you want to use',
agent: 'Agent',
addAgent: 'Add agent',
diff --git a/web/src/locales/zh.ts b/web/src/locales/zh.ts
index 126dc0a6fd..ca57c72b67 100644
--- a/web/src/locales/zh.ts
+++ b/web/src/locales/zh.ts
@@ -2940,8 +2940,6 @@ NER:使用 spaCy NER 和基于规则的关键词提取来抽取 Entities 和 R
query: '查询变量',
queryRequired: '查询变量是必填项',
queryTip: '选择您想要使用的变量',
- documentIds: '文档 ID 列表',
- documentIdsTip: '可选,限制检索范围的文档 ID 列表,支持引用上游变量',
agent: '智能体',
addAgent: '添加智能体',
agentDescription: '构建具备推理、工具调用和多智能体协同的智能体组件。',
diff --git a/web/src/pages/agent/constant/index.tsx b/web/src/pages/agent/constant/index.tsx
index 57a08589d1..ca101cd64f 100644
--- a/web/src/pages/agent/constant/index.tsx
+++ b/web/src/pages/agent/constant/index.tsx
@@ -94,7 +94,6 @@ export const initialRetrievalValues = {
...initialKeywordsSimilarityWeightValue,
cross_languages: [],
retrieval_from: RetrievalFrom.Dataset,
- document_ids: '',
outputs: {
formalized_content: {
type: 'string',
diff --git a/web/src/pages/agent/form/retrieval-form/next.tsx b/web/src/pages/agent/form/retrieval-form/next.tsx
index 11fd2d99ee..481886d855 100644
--- a/web/src/pages/agent/form/retrieval-form/next.tsx
+++ b/web/src/pages/agent/form/retrieval-form/next.tsx
@@ -55,7 +55,6 @@ export const RetrievalPartialSchema = {
memory_ids: z.array(z.string()).optional(),
retrieval_from: z.string(),
user_id: z.string().optional(),
- document_ids: z.string().optional(),
};
export const FormSchema = z
@@ -134,20 +133,6 @@ export function useHideKnowledgeGraphField(form: UseFormReturn<any>) {
return retrievalFrom === RetrievalFrom.Memory;
}
-export function DocumentIdsFormField() {
- const { t } = useTranslation();
-
- return (
- <RAGFlowFormItem
- name="document_ids"
- label={t('flow.documentIds')}
- tooltip={t('flow.documentIdsTip')}
- >
- <PromptEditor multiLine={false} showToolbar={false}></PromptEditor>
- </RAGFlowFormItem>
- );
-}
-
function RetrievalForm({ node }: INextOperatorForm) {
const { t } = useTranslation();
const ownerTenantId = useOwnerTenantId();
@@ -190,7 +175,6 @@ function RetrievalForm({ node }: INextOperatorForm) {
<RAGFlowFormItem name="query" label={t('flow.query')}>
<PromptEditor></PromptEditor>
</RAGFlowFormItem>
- <DocumentIdsFormField></DocumentIdsFormField>
<MemoryDatasetForm></MemoryDatasetForm>
<Collapse defaultOpen title={<div>{t('flow.advancedSettings')}</div>}>
<section className="space-y-5">
diff --git a/web/src/pages/agent/form/tool-form/retrieval-form/index.tsx b/web/src/pages/agent/form/tool-form/retrieval-form/index.tsx
index 8b1338bb6b..624102ec6d 100644
--- a/web/src/pages/agent/form/tool-form/retrieval-form/index.tsx
+++ b/web/src/pages/agent/form/tool-form/retrieval-form/index.tsx
@@ -22,7 +22,6 @@ import { useOwnerTenantId } from '../../../context';
import { DescriptionField } from '../../components/description-field';
import { FormWrapper } from '../../components/form-wrapper';
import {
- DocumentIdsFormField,
MemoryDatasetForm,
RetrievalPartialSchema,
useHideKnowledgeGraphField,
@@ -61,7 +60,6 @@ const RetrievalForm = () => {
<Form {...form}>
<FormWrapper>
<DescriptionField></DescriptionField>
- <DocumentIdsFormField></DocumentIdsFormField>
<MemoryDatasetForm></MemoryDatasetForm>
<Collapse defaultOpen title={<div>{t('flow.advancedSettings')}</div>}>
<FormContainer>