kektordb MCP工具

Installation

As a Server (Docker)

docker run -p 9091:9091 -p 9092:9092 -v $(pwd)/data:/data sanonone/kektordb:latest

Use Cases

🚀 Quick Start (Python)

This example demonstrates building an AI Agent with Memory using sessions and cognitive features.

1. Install Client:

    pip install kektordb-client sentence-transformers
    

1. Run the script:

    from kektordb_client import KektorDBClient
    from kektordb_client.cognitive import CognitiveSession

    # 1. Initialize
    client = KektorDBClient(port=9091)
    index = "agent_memory"

    # 2. Create index with memory enabled (30 day half-life)
    try:
        client.delete_index(index)
    except:
        pass
    client.vcreate(index, metric="cosine", text_language="english")

    # 3. Start a session (e.g., user conversation)
    with CognitiveSession(client, index, user_id="user_123") as session:
        # Save memories linked to this conversation
        session.save_memory(
            "User's name is Marco and he prefers concise answers",
            layer="episodic",
            tags=["user_preference"]
        )
        session.save_memory(
            "Marco is working on a Go project called KektorDB",
            layer="semantic",
            tags=["project", "go"]
        )
        session.save_memory(
            "Marco asked how to implement RAG. Explained vector search + graph traversal.",
            layer="episodic"
        )

    # 4. Later: Retrieve user profile
    profile = client.get_user_profile("user_123", index)
    print(f"Communication style: {profile.get('communication_style', 'N/A')}")
    print(f"Expertise areas: {profile.get('expertise_areas', [])}")

    # 5. Search memories
    results = client.vsearch(
        index,
        query_vector=[0.1, 0.2, 0.3, 0.4],  # Embed your query
        k=3,
        filter_str="tags ? 'project'"
    )
    print(f"Found {len(results)} relevant memories")

    # 6. Check belief confidence of a memory
    belief = client.vbelief_state(index, query="Marco's project name")
    print(f"Confidence: {belief['confidence']:.2f}, State: {belief['state']}")

    # 7. Evolve a memory when information changes
    new_id = client.vevolve(
        index,
        old_id="memory_001",
        reason="User updated their preference",
        new_content="Marco now prefers detailed explanations with examples"
    )
    print(f"Memory evolved to: {new_id['new_id']}")
    

👉 Read the Full Documentation for all available endpoints and features.

---

With custom options

./kektordb -http-addr :9091 -save "30 500" -log-level debug ```

Command-line flags: -http-addr: HTTP server address (default: :9091) -aof-path: Path to persistence file (default: kektordb.aof) -save: Auto-snapshot policy "seconds changes" (default: "60 1000", empty to disable) -auth-token: API authentication token -log-level: Logging level (debug, info, warn, error) -enable-proxy: Enable AI Gateway/Proxy -proxy-config: Path to proxy configuration file -vectorizers-config: Path to vectorizers configuration file -mcp: Run as MCP Server (Stdio) --embedder: Embedder mode: auto, ollama, openai, local --embedder-model: Directory with ONNX model + tokenizer (local mode) --tools: MCP tool profile: agent, admin, all --tui: Launch terminal dashboard (experimental) --cognitive-config: Path to cognitive YAML config (enables Gardener)

Compatibility Note: All development and testing were performed on Linux (x86_64). Pure Go builds are expected to work on Windows/Mac/ARM.

---

Built-in Embedding (Optional)

KektorDB includes an optional built-in ONNX embedder (all-MiniLM-L6-v2, 384 dimensions) powered by Rust/Candle for zero-config local embeddings -- no Ollama required.

Build with Rust support:

make build-rust-native    # requires protoc (auto-downloaded by Makefile)
make run-rust

Or manually:

cd native/compute && cargo build --release
CGO_LDFLAGS="-L$(pwd)/native/compute/target/release" go build -tags rust ./cmd/kektordb/

The ONNX model (~90 MB) is downloaded automatically from HuggingFace on first launch with SHA256 verification.

---

Zero-Code RAG (Open WebUI Integration)

<p align="center"> <img src="docs/images/kektordb-rag-demo.gif" alt="KektorDB RAG Demo" width="800"> </p>

KektorDB can function as a smart middleware between your Chat UI and your LLM. It intercepts requests, performs retrieval, and injects context automatically.

Architecture: Open WebUI -> KektorDB Proxy (9092) -> Ollama / LocalAI (11434)

How to set it up:

1. Configure vectorizers.yaml to point to your documents and enable Entity Extraction. 2. Configure proxy.yaml to point to your Local LLM (Ollama) or OpenAI. 3. Run KektorDB with the proxy enabled:

    ./kektordb -vectorizers-config='vectorizers.yaml' -enable-proxy -proxy-config='proxy.yaml'
    

👉 Read the Full Guide: Building a Fast RAG System with Open WebUI

---

🦜 Integration with LangChain

KektorDB includes a built-in wrapper for LangChain Python, allowing you to plug it directly into your existing AI pipelines.

from kektordb_client.langchain import KektorVectorStore

---