ApeRAG架构分析

1
2
3
4

git clone https://github.com/apecloud/ApeRAG.git
cd ApeRAG
cp envs/env.template .env
docker-compose up -d --pull always

 1
 2
 3
 4
 5
 6
 7
 8
 9
10

{
  "mcpServers": {
    "aperag-mcp": {
      "url": "https://localhost:8000/mcp/",
      "headers": {
        "Authorization": "Bearer your-api-key-here"
      }
    }
  }
}

1
2

git clone https://github.com/apecloud/ApeRAG.git
cd ApeRAG

1
2
3
4
5
6
7
8

cd deploy/databases/

bash ./01-prepare.sh # 安装 KubeBlocks
bash ./02-install-database.sh # 部署 PostgreSQL、Redis、Qdrant、Elasticsearch

kubectl get pods -n default # 检查部署情况

cd ../../

1
2
3

helm install aperag ./deploy/aperag --namespace default --create-namespace # 部署 Ape 服务

kubectl get pods -n default -l app.kubernetes.io/instance=aperag # 检查服务状态

1
2

kubectl port-forward svc/aperag-fronted 3000:3000 -n default
kubectl port-forward svc/aperag-api 8000:8000 -n default

 1
 2
 3
 4
 5
 6
 7
 8
 9
10
11

Summarize this text chunk concisely. Requirements:
1. Use the same language as the original text for the summary
2. Keep it within 1-2 sentences
3. Extract only the most important core information
4. Stay objective and accurate, do not add content not present in the original text
5. Output ONLY the summary content, no additional text, explanations, or formatting

Text content:
{text}

Summary:

 1
 2
 3
 4
 5
 6
 7
 8
 9
10
11
12

Generate a concise summary of this document. Requirements:
1. Use the same language as the original text for the summary
2. Keep it within 2-3 sentences
3. Summarize the main topic and key insights of the document
4. Stay objective and accurate, do not add content not present in the original text
5. If it's a technical document, highlight the technical points
6. Output ONLY the summary content, no additional text, explanations, or formatting

Document content:
{text}

Summary:

 1
 2
 3
 4
 5
 6
 7
 8
 9
10
11
12
13

Combine these section summaries into a comprehensive final document summary. Requirements:
1. Use the same language as the original summaries for the final summary
2. Keep it within 3-4 sentences
3. Integrate the core content from all sections into a coherent overall summary
4. Highlight the main topic and most important insights of the document
5. Maintain logical clarity and avoid repetitive content
6. If technical content is involved, maintain accuracy of technical terminology
7. Output ONLY the final summary content, no additional text, explanations, or formatting

Section summaries:
{combined_summaries}

Final summary:

 1
 2
 3
 4
 5
 6
 7
 8
 9
10
11
12
13
14
15
16
17
18
19
20

    Analyze the provided image and extract its content with high fidelity. Follow these instructions precisely and use Markdown for formatting your entire response. Do not include any introductory or conversational text.

1.  **Overall Summary:**
    *   Provide a brief, one-paragraph overview of the image's main subject, setting, and any depicted activities.

2.  **Detailed Text Extraction:**
    *   Extract all text from the image, preserving the original language. Do not translate.
    *   **Crucially, maintain the visual reading order.** For multi-column layouts, process the text column by column (e.g., left column top-to-bottom, then right column top-to-bottom).
    *   **Exclude headers and footers:** Do not extract repetitive content from the top (headers) or bottom (footers) of the page, such as page numbers, book titles, or chapter names.
    *   Replicate the original formatting using Markdown as much as possible (e.g., headings, lists, bold/italic text).
    *   For mathematical formulas or equations, represent them using LaTeX syntax (e.g., `$$...$$` for block equations, `$...$` for inline equations).
    *   For tables, reproduce them accurately using GitHub Flavored Markdown (GFM) table syntax.

3.  **Chart/Graph Analysis:**
    *   If the image contains charts, graphs, or complex tables, identify their type (e.g., bar chart, line graph, pie chart).
    *   Explain the data presented, including axes, labels, and legends.
    *   Summarize the key insights, trends, or comparisons revealed by the data.

4.  **Object and Scene Recognition:**
    *   List all significant objects, entities, and scene elements visible in the image.

1
2
3
4
5

class VectorSearchInput(BaseModel):
    top_k: int = Field(5, description="Number of top results to return")
    similarity_threshold: float = Field(0.2, description="Similarity threshold for vector search")
    collection_ids: Optional[List[str]] = Field(default_factory=list, description="Collection IDs")
    chat_id: Optional[str] = Field(None, description="Chat ID to filter chat documents")

1
2

class VectorSearchOutput(BaseModel):
    docs: List[DocumentWithScore]

1
2
3
4
5
6
7
8

class FulltextSearchInput(BaseModel):
    query: str = Field(..., description="User's question or query")
    top_k: int = Field(5, description="Number of top results to return")
    collection_ids: Optional[List[str]] = Field(default_factory=list, description="Collection IDs")
    keywords: Optional[List[str]] = Field(
        default_factory=list, description="Custom keywords to use for fulltext search"
    )
    chat_id: Optional[str] = Field(None, description="Chat ID to filter chat documents")

1
2

class FulltextSearchOutput(BaseModel):
    docs: List[DocumentWithScore]

1
2
3
4

class SummarySearchInput(BaseModel):
    top_k: int = Field(5, description="Number of top results to return")
    similarity_threshold: float = Field(0.2, description="Similarity threshold for summary search")
    collection_ids: Optional[List[str]] = Field(default_factory=list, description="Collection IDs")

1
2

class SummarySearchOutput(BaseModel):
    docs: List[DocumentWithScore]

1
2
3

class GraphSearchInput(BaseModel):
    top_k: int = Field(5, description="Number of top results to return")
    collection_ids: Optional[list[str]] = Field(default_factory=list, description="Collection IDs")

1
2

class GraphSearchOutput(BaseModel):
    docs: List[DocumentWithScore]

1
2
3
4

class VisionSearchInput(BaseModel):
    top_k: int = Field(5, description="Number of top results to return")
    similarity_threshold: float = Field(0.2, description="Similarity threshold for vision search")
    collection_ids: Optional[List[str]] = Field(default_factory=list, description="Collection IDs")

1
2

class VisionSearchOutput(BaseModel):
    docs: List[DocumentWithScore]

 1
 2
 3
 4
 5
 6
 7
 8
 9
10
11
12
13
14
15
16
17

class MergeInput(BaseModel):
    merge_strategy: str = Field("union", description="How to merge results")
    deduplicate: bool = Field(True, description="Whether to deduplicate merged results")
    vector_search_docs: Optional[List[DocumentWithScore]] = Field(
        default_factory=list, description="Vector search docs"
    )
    fulltext_search_docs: Optional[List[DocumentWithScore]] = Field(
        default_factory=list, description="Fulltext search docs"
    )
    graph_search_docs: Optional[List[DocumentWithScore]] = Field(default_factory=list, description="Graph search docs")
    summary_search_docs: Optional[List[DocumentWithScore]] = Field(
        default_factory=list, description="Summary search docs"
    )
    vision_search_docs: Optional[List[DocumentWithScore]] = Field(
        default_factory=list, description="Vision search docs"
    )
    qa_search_docs: Optional[List[DocumentWithScore]] = Field(default_factory=list, description="Q&A search docs")

1
2

class MergeOutput(BaseModel):
    docs: List[DocumentWithScore]

1
2
3
4
5
6
7
8

class RerankInput(BaseModel):
    use_rerank_service: bool = Field(default=True, description="Whether to use rerank service or fallback strategy")
    model: Optional[str] = Field(default=None, description="Rerank model name")
    model_service_provider: Optional[str] = Field(default=None, description="Model service provider")
    custom_llm_provider: Optional[str] = Field(
        default=None, description="Custom LLM provider (e.g., 'jina_ai', 'openai')"
    )
    docs: List[DocumentWithScore]

1
2

class RerankOutput(BaseModel):
    docs: List[DocumentWithScore]