开源向量数据库，可安装并查询

停止租用你的智能，用 AnythingLLM 拥有它，获得强大的本地优先 agent 体验所需的一切。

测试发现：The AnythingLLM monorepo installs and builds cleanly; the server boots on port 3001 responding to API calls; the collector boots on port 8888; all 1,119 unit tests across server (725) and collector (394) pass without errors. 20 分钟。

面向机器人和 agent 的 AI 搜索引擎与 MCP 服务器。单个 Rust 二进制文件即可索引代码、文档、日志、JSONL/CSV 和 PDF，用于 BM25 全文、向量和混合搜索、无需 grep 的代码搜索、agent 记忆与 RAG。兼容 Elasticsearch 的 API；Corpus Hub...

测试发现：XERJ v1.0.0-rc.77 built from source in 2m06s (quick profile), serving ES-compat API on :9200 with autoindex and MCP server working; test suites pass except 2 IPv6 bind failures specific to this container. 36 分钟。

LangChain4j 是一个地道的开源 Java 库，用于在 JVM 上构建由 LLM 驱动的应用程序。它为流行的 LLM 提供商和向量存储提供统一 API，并简化了工具调用（包括 MCP 支持）、agent 和 RAG 的实现。它与 Quarkus 和 Spring Boot 等企业级 Java 框架无缝集成。

测试发现：LangChain4j project compiles and 1724 unit tests pass with Java 17 (Adoptium Temurin-17.0.21+9) and Maven wrapper 3.9.16. 32 分钟。

用于在 JavaScript、Go、Dart 和 Python 中构建 agent 应用的开源框架，由 Google 构建并用于生产环境。

测试发现：Genkit JS SDK core + genkit package, all 23 JS plugins, genkit-tools (cli, common, telemetry-server), and Python SDK install, build, and run test suites, with the genkit CLI printing version 1.44.1 and help text. 26 分钟。

用于 LLM 应用的本地持久化记忆存储，包括 claude desktop、github copilot、codex、antigravity 等。

测试发现：The LongMemory Hydrograph TypeScript package builds, its CLI (init, ingest, recall, doctor) works, the library API (createMemory, ingest, recall) returns correct results, the HTTP server starts on port 7331 and responds 200 on /health, and the CI benchmark suite passes all 11 smoke-test gates with 100% retrieval... 9 分钟。

CrateDB 是一个分布式、可扩展的 SQL 数据库，用于近乎实时地存储和分析海量数据，即使面对复杂查询也能胜任。它与 PostgreSQL 兼容，并基于 Lucene。

测试发现：CrateDB 6.5.0 compiles, installs, launches as a single-node cluster, accepts HTTP SQL queries, and returns correct query results. 27 分钟。

面向 AI 原生应用的分布式向量搜索

测试发现：Vearch v3.5.9 binary (81MB) and gamma engine (.so) built from source, server starts with master+router+PS on localhost:8817/9001, REST API responds, db/create/list works, cluster health/stats/members endpoints return valid JSON, tests pass basic cluster operations but space creation times out due to PS heartbeat... 59 分钟。

面向 AI agent 流水线（LangGraph、CrewAI、AutoGen）和 Claude 的开源持久化内存。REST API + 知识图谱 + 自主整合。

测试发现：mcp-memory-service 11.13.0 installed from source via pip install -e '.[dev]', the uvicorn server starts on 127.0.0.1:9791 (HTTP) with sqlite_vec backend, health endpoint returns 200, memory CRUD operations work end to end, and the full test suite passes except for one benchmark test that fails due to hash-embedding... 20 分钟。

AI 原生 HTAP 数据库，具备 Git-for-Data 和内置向量搜索，作为智能 agent 和应用的数据和记忆骨干。

测试发现：MatrixOne server built from source (mo-service, 275MB binary), launched on port 6001, accepting MySQL protocol connections, queries returning correct results including data manipulation operations. 34 分钟。

SeekStorm：向量与词法搜索，进程内库与多租户服务器，使用 Rust 编写。

测试发现：Workspace builds cleanly with Rust 1.99.0, all 45 seekstorm library integration tests and all 7 seekstorm_client E2E tests pass, and the release seekstorm_server binary serves the live endpoint (HTTP 200) and completes the API key creation, index creation, document indexing, commit, and query flow returning the... 63 分钟。

LLPhant：一个全面的 PHP 生成式 AI 框架，使用 OpenAI GPT 4。灵感来自 Langchain。

测试发现：Static PHP 8.2.32 and Composer 2.7.9 installed; all 119 Composer dependencies resolved; 223 unit tests pass in 0.3s; PHPStan static analysis passes with 0 errors; Laravel Pint style check passes on 294 files. 6 分钟。

面向构建者和黑客的开放文档智能平台 - 面向 agent 世界的 DMS

测试发现：Python dependencies installed in virtual environment, embedded PostgreSQL 18.6 provisioned and running, Django app boots with all migrations applied, and 613+ tests pass verifying admin, agents, annotations, authentication, analysis pipeline, badge system, base services, and artifact service against a real PostgreSQL... 77 分钟。

Milvus Vector Database 的 Python SDK

测试发现：PyMilvus SDK installed via uv sync, imports successfully, get_commit returns '290d76f', and the full unit test suite passes 5255/5255 with 3 skipped. 36 分钟。

Qdrant 向量搜索引擎的 Python 客户端

测试发现：The qdrant-client Python library installed without errors. All local-mode tests pass. When a Qdrant server v1.19.1 runs in the same session, all REST and gRPC integration tests pass. The in-memory client creates collections, upserts points, and searches vectors successfully. 32 分钟。

基于邻域图和树的高维数据最近邻搜索

测试发现：NGT v2.8.1 C++ library, command-line tool, and four sample programs build and run successfully; the CLI creates indexes, appends data, searches with L2/cosine/hamming distance, and reports index statistics; the Python ctypes binding (ngt.base) inserts 5000 128-d vectors and returns correct nearest-neighbor results. 11 分钟。

Infinispan 是一个开源数据网格平台，也是一个高度可扩展的 NoSQL 云数据存储。

测试发现：Infinispan server 16.3.0-SNAPSHOT starts on 127.0.0.1:11222, responds to REST API requests for server info and cache operations, and the core module's SimpleCacheTest passes all 115 test cases. 21 分钟。

本地优先的语义代码搜索引擎

测试发现：SeaGOAT v1.2.0 installs via poetry, starts a local server, responds to status queries, and returns correct semantic search results for a test repository. 190 of 204 tests pass; 12 snapshot tests fail due to ONNX model processing timeout on this CPU. 30 分钟。

可嵌入的 Go 向量数据库，提供类似 Chroma 的接口，且零第三方依赖项。内存中运行，可选持久化。

测试发现：Built and tested chromem-go, all tests passing and benchmarks completing with zero third-party dependencies. 4 分钟。

Lite & Super-fast 重新排序，用于搜索与检索流水线。支持基于 LLM、cross-encoder 等的最新 Listwise 和 Pairwise 重排序。由 Prithivi Da 创建，欢迎提交 PR 与合作。

测试发现：FlashRank 0.2.9 is installed in .venv and the README reranking example runs end to end against the real TinyBERT model, printing the same ranked output as the README. 2 分钟。

基于 ID 的 RAG FastAPI：与 Langchain 和 PostgreSQL/pgvector 的集成

测试发现：Dependencies installed in venv, all 303 non-integration tests pass after adding OPENAI_API_KEY mock to conftest.py; 6 tests skip due to missing pandoc, 35 integration tests require Docker. 4 分钟。

OpenSwarm：由 Claude Code CLI 驱动的自主 AI 开发团队编排器。Discord 控制，Linear 集成，认知记忆。

测试发现：npm install, npm run build, and 493 of 495 test file suites pass (6746/6757 individual tests). The project CLI runs and all major subsystems (verify, sandbox, worktree, memory, diagnostics) are operational. 48 分钟。

使用 Rust 和 DataFusion 学习向量搜索；C++/BusTub 路线已弃用。

测试发现：The workspace builds and all 87 solution tests pass: in-memory vector indexes (Flat, IVFFlat, IVF-PQ, NSW, HNSW) are implemented in the core crate, integrated with DataFusion via SQL optimizer rules and MemTable attachments, and validated by sqllogictest-based end-to-end SQL tests. The 5 SIFT1M smoke tests are skipped... 15 分钟。

RAGLight 是一个用于检索增强生成（RAG）的模块化框架。它让接入不同的 LLMs、嵌入和向量存储变得简单，并且现在包含无缝的 MCP 集成，以连接外部工具和数据源。

测试发现：RAGLight 3.4.7 is fully installed in a venv at /tmp/venv with all extras (chroma, qdrant, langfuse). The CLI, FastAPI REST API server, and all 75 unit tests (embeddings, LLM, vector store, API, document processing, scrapper, config) all pass and function correctly. The package is importable from Python under the venv. 6 分钟。

NextPlaid，ColGREP：多向量搜索，从数据库到 coding agents。

测试发现：The Rust workspace compiles, all 1062 unit tests pass, both binaries respond to help/version, NextPlaid API serves CRUD endpoints (index create/add documents/search returning real scores), and ColGREP builds indexes and performs semantic code search with ONNX model inference. 14 分钟。

加密、完全离线的 agent 记忆。一键安装，带记忆地图的 GUI，支持所有操作系统和 agent。卓越的记忆创建、存储和检索。

测试发现：Compartment v4.10.1 installs and runs in Python 3.12.3 venv; vault creates with 6664 seeded starter memories; MCP stdio server responds to initialize/tools/list/store/search/forget; test suite passes 975+ tests with 32 platform skips and 1 deselected real_launchctl test. 76 分钟。

本地优先的 AI agent 记忆，包含知识图谱、本体约束和可追溯推理，全部在一个 SQLite 文件中。

测试发现：CortexDB v2.113.0 builds from source, all 25 test packages pass, and the gRPC+REST server starts on lexical mode (no API keys) serving health, info, knowledge, memory, graph, and query endpoints with no errors. 15 分钟。

Qdrant Vector Database 的 JavaScript/Typescript SDK

测试发现：All 112 tests pass across all 4 packages (js-client-rest unit: 28, js-client-rest integration: 25, js-client-grpc unit: 42, js-client-grpc integration: 17) against Qdrant 1.19.2 running locally on port 6333. 5 分钟。

快速、基于 SQL 的进程内向量搜索，支持任何语言，带有 SQLite 驱动程序

测试发现：vectorlite.so builds and loads as a SQLite extension; all 35 C++ unit tests and all 112 Python integration tests pass on Ubuntu 24.04 x86_64 with vcpkg-managed dependencies. 8 分钟。

轻量级最近邻搜索，支持灵活的多种后端

测试发现：Package installs and runs in Python 3.12; BASIC, FAISS, PYNNDESCENT, USEARCH, and VOYAGER backends work correctly; HNSW and ANNOY backends require python3-dev to compile from source. 12 分钟。

面向 AI agent 的本地无 LLM 记忆。一个单独的离线 Rust 二进制程序，确定性且可审计，它从使用中学习、遗忘无关信息并强化重要内容。无云，无 API 密钥。

测试发现：The Shodh-Memory v0.2.0 pre-built server binary stores and recalls memories via REST API on localhost:3031, the MCP server TypeScript test suite passes 216/216, and the Python client test suite passes 106/106. 21 分钟。

面向 Claude Code 的本地 RAG MCP 服务器：混合搜索（语义 + BM25）、交叉编码器重排序、13 个 MCP 工具、20 个格式解析器。零外部服务器，零 API 密钥。

测试发现：knowledge-rag v4.9.0 installed from source in a Python 3.12 venv; all 779 automated tests pass; the MCP server starts and responds to initialize and tools/list MCP protocol messages over stdio. 5 分钟。

AI 劳动力之家：多 agent 系统、AI agent 与工具

测试发现：The relevanceai SDK version 10.2.1 is installed and all 46 unit tests pass, covering client initialization, agents, tools, tasks, knowledge, and OAuth resource classes via mocked HTTP responses. 1 分钟。

高性能多向量搜索引擎

测试发现：fast-plaid 1.7.0 builds, installs from a patched wheel, and passes all 83 tests: create, search, update, delete, metadata filtering, freeze/unfreeze, single-copy storage, legacy argument compatibility, and chunk-budget verification on CPU with torch 2.7.0. 31 分钟。

适用于任何 AI agent 的长期记忆。单个 Rust 二进制文件，离线，在 LongMemEval-S 上 recall@5 达到 98.2%。免费自托管或使用 Uteke Cloud。

测试发现：Uteke v0.18.2 builds from source, stores and semantically recalls memories via CLI (remember/recall), serves HTTP API on port 18768 (health/remember/recall all 200 OK with correct JSON), and the MCP server responds to initialize, all 705 workspace tests pass with 0 failures. 14 分钟。

Radient 可将多种数据类型（不仅仅是文本）转换为向量，用于相似性搜索、RAG、回归分析等场景。

测试发现：Installs under virtualenv. All 5 vectorizer backends produce correct embeddings. GANN C extension compiles and searches. Workflows run. ffmpeg demux produces decodable media. accelerate() broken on modern torch/transformers. 25 分钟。

用于 Pinecone 向量数据库的官方 TypeScript/Node 客户端

测试发现：SDK installs, builds, and passes all 1460 unit tests and 48 mocked smoke tests on Node 22.14.0. The dist/ folder contains 30 files with all expected exports (Pinecone, Index, Documents, Assistants, Inference, etc.). No errors remain. 3 分钟。

一个简单且易于修改的向量数据库

测试发现：The nano-vectordb package installs in a local venv, all 7 tests pass, and real upsert/query/get/delete/filter/save-reload and multi-tenant persistence operations all worked in this container. 1 分钟。

Chroma 向量数据库的 Go 客户端

测试发现：make build succeeds. 1975 V2 API tests pass, default_ef embedding function produces real 384-dim embeddings via ONNX Runtime, ORT and BM25 embedding function tests all pass. 20 分钟。

这并非又一个 ChatGPT 包装器，而是一个本地优先的工具，可将你的笔记、网页、PDF 和音视频转化为个人知识图谱，完全在你的机器上运行。

测试发现：Backend API serves 30 endpoints with seed graph data, all 131 automated tests pass, and the React frontend builds and starts. 4 分钟。

一个为 Hermes Agent 设计的 7 层内存操作系统：持久化内存（使用 Qdrant）、结构化事实、结构召回、自动整理的知识库以及精准上下文注入。可在本地运行，支持任何 LLM 提供商。

测试发现：Python dependencies install cleanly in a venv, all 7 icarus module imports succeed, all 3 standalone test suites pass with 0 failures, SQLite databases are initialized with correct FTS5 schema, Icarus plugin tools produce valid output, and the Docker-compose file, worker services, and cron scripts are syntactically... 8 分钟。

轻量级、极速的进程内向量数据库

测试发现：zvec v0.8.0 installed via pip prebuilt wheel on Python 3.12.3 (x86_64, Linux). Vector ANN query, FTS query, collection stats, and all core operations verified working. C++ static library (libzvec.a, 15MB) built from source. Python wheel build from source fails without python3-dev headers. 30 分钟。

PageIndex：面向无向量、基于推理的 RAG 的文档索引

测试发现：PageIndex SDK v0.2.10 installed in a venv, all 796 unit tests pass, the CLI parses correctly, and the package imports without error. 3 分钟。

Milvus 是一个高性能、云原生的向量数据库，专为可扩展的向量 ANN 搜索而构建。

测试发现：pymilvus 3.0.2 with milvus-lite 3.2.1 runs embedded vector database operations (create, insert 1000 vectors, search, query, delete, drop) in the container. Go toolchain 1.26.6, Rust 1.84, and Conan 2.25.1 are installed. The C++ core cannot build without root for system library packages. 31 分钟。

HelixDB 是一个 OLTP 图数据库，具有原生向量和全文搜索功能，使用 Rust 构建于对象存储之上。

测试发现：The HelixDB workspace builds and all crate test suites pass. The server binary starts and responds to HTTP POST /v2/query with HTTP 200. The CLI binary prints version 3.3.0 and help output. 51 分钟。

为 LLM 应用构建的 AI 原生数据库，提供对稠密向量、稀疏向量、张量（多向量）和全文的极快混合搜索。

测试发现：Infinity v0.7.3 nightly (x86_64-v2) standalone server starts and accepts Thrift connections on port 23817 and HTTP on port 23820; Python SDK can create tables, insert vectors, run dense vector search queries, and return results via Polars DataFrames. 12 分钟。

ArcadeDB 多模型数据库，一个支持 SQL、Cypher、Gremlin、HTTP/JSON、MongoDB 和 Redis 的 DBMS。ArcadeDB 是 OrientDB（首个多模型 DBMS）的概念分支。ArcadeDB 支持向量嵌入。

测试发现：ArcadeDB 26.10.1-SNAPSHOT is fully built and functional: the server starts in development mode listening on port 2480, serving the Studio web UI (HTTP 200), and the engine test suite passes ~14624 tests with 0 errors and 1 pre-existing flaky timeout test that fails under heap contention during concurrent runs. 85 分钟。

向量数据库的 benchmark

测试发现：vectordb-bench installed from source in venv, 247 of 256 unit tests pass, CLI and REST API both operable. 22 分钟。

用于 RaBitQ 向量量化和近似最近邻搜索的官方库，使用 C++ 和 Python。

测试发现：RaBitQ Library v0.4.0 Python bindings and C++ library built from source; 395/395 Python tests and 187/187 C++ tests pass natively on x86_64 with runtime AVX2/AVX-512 dispatch. 11 分钟。

PHP 生态系统中用于构建生产级 AI 驱动应用的 Agent 框架。将组件（LLMs、工具、向量数据库、记忆）连接到与你的数据和 UI 交互的 agents。

测试发现：Composer install with all 145 dependencies succeeded; vendor/bin/phpunit reports 1373 tests, 5460 assertions passed, 0 failures, 0 errors, 78 skipped (external services); vendor/bin/phpstan analyse reports OK, 0 errors; bin/neuron CLI reports help with 7 commands. 33 分钟。

Cognee 是面向 agent 的开源 AI 记忆平台。使用小模型即可免费为你的 AI agent 提供持久长期记忆。

测试发现：cognee 1.6.1 installs, builds, starts, and runs: the local LLM-free remember/recall pipeline produces search results from ingested text, the CLI demo loads a bundled knowledge graph, the FastAPI server answers HTTP 200 on /health, and 388 unit tests pass across 5 test suites with zero failures. 33 分钟。

面向开发者的开源嵌入式检索库，用于多模态人工智能。搜索更多；管理更少。

测试发现：Rust core crate lancedb compiles and passes all 73 integration tests; Python bindings install, import, and pass 1368 tests; the local (in-process) embedding store works end-to-end with vector search queries returning correct results. 84 分钟。

用于语义搜索、LLM 编排和语言模型工作流的一体化 AI 框架

测试发现：txtai 9.14.0 installed in venv, all core embeddings/graph/workflow/agent/cloud/vector/scoring tests pass, API responds to search/index/count endpoints with real model inference, ~470 automated tests pass across 30 test suites. 83 分钟。

用于 Claude Code 的代码搜索 MCP。将整个代码库作为任何 coding agent 的上下文。

测试发现：Install, build, typecheck, lint config, and all 35 tests pass across the monorepo. The core indexing pipeline runs end-to-end with mock providers. The MCP server starts but requires real OpenAI API key and Milvus credentials for production use. 15 分钟。

MineContext 是您的主动式上下文感知 AI 伙伴（Context-Engineering+ChatGPT Pulse）

测试发现：MineContext server starts successfully, serves all 12 HTTP routes (GET and POST), initializes ChromaDB vector store and SQLite document store, and the CLI help command prints usage information. 18 分钟。

使用本地向量数据库为 coding agents 提供持久记忆的 OpenCode 插件

测试发现：Install, TypeScript compilation, all 767 tests, full build (TS + web UI), and the web server all succeed on Linux/amd64 with Bun 1.4.3. The package.json was patched to use bun --bun for the web:build script, because system Node 18 cannot run the project's modern Svelte/Vite toolchain. 13 分钟。

认识一下 Ava，WhatsApp Agent

测试发现：uv sync installs all dependencies, 30 Python source files compile, LangGraph workflow graph compiles with 9 nodes, FastAPI WhatsApp webhook answers GET verification with HTTP 200, Chainlit chat interface starts on port 8000 and serves HTTP 200, mock API keys produce 401 when POSTing messages as expected since real... 19 分钟。

ChatWeb 可以爬取网页、读取 PDF、DOCX、TXT 文件，提取主要内容，然后根据内容回答你的问题，或总结关键要点。

测试发现：ChatWeb runs in all three modes (console, API, webui) behind a mock OpenAI API. The /upload_file, /summary, and /answer endpoints return HTTP 200 with correct JSON schemas; the console mode accepts file paths and /help commands; the webui Gradio interface binds on port 7860 and returns HTTP 200. 15 分钟。

AI 助手的长期记忆。图 + 向量存储，可跨会话召回决策、关系和上下文。

测试发现：In this container: make install succeeds, python app.py boots and serves /health, the unit suite passes 664 tests, and authenticated store/recall/query/delete flows work over HTTP against the schema-compatible in-memory mock. 15 分钟。

测试发现：The crate builds, all 5 tests pass, the HTTP server answers 200 on /ping and returns isError:false for all 5 MCP tools against a local schema-correct mock Elasticsearch, and stdio mode completes the MCP initialize handshake. 11 分钟。

一个 NodeJS RAG 框架，可轻松处理 LLM 和嵌入。

测试发现：EmbedJs monorepo installs and builds; RAG pipeline (loaders -> embeddings -> HNSWDb -> similarity search) works with mocked embeddings. 11 分钟。

一个简单的 VoiceRAG 模式示例实现，用于使用 Azure AI Search 和 Azure OpenAI 的 gpt-4o-realtime-preview 模型，通过 RAG 支持交互式语音生成式 AI 体验。

测试发现：Backend server starts on port 8765, serves built frontend (HTML, JS, CSS, favicon all HTTP 200), WebSocket endpoint /realtime accepts client connections and routes to Azure OpenAI, all Python imports compile, Ruff lint passes, frontend TypeScript builds cleanly via Vite. 3 分钟。

📊 电商数仓智能问数 AI Agent，最适合用于系统学习 LangGraph 的实战项目：基于 LangGraph、FastAPI、Qdrant、Elasticsearch、MySQL 与 React，完整实现元数据知识库、混合检索、自然语言生成 NL2SQL 生成校验、SQL...

测试发现：With schema-correct mocks for the mandatory external services, the uvicorn-served app answers POST /api/query with a complete SSE pipeline and returns SQL aggregate result sales=10088 for '统计华北地区销售额'; uv sync installs all dependencies and the frontend pnpm build completes. 31 分钟。

基于 Upstash Vector 的维基百科语义搜索

测试发现：pnpm install and next build succeed; next start serves / with HTTP 200 and the full chat/search pipeline runs end-to-end against local schema-correct Upstash/OpenAI mocks, with retrieval context and streaming answers verified via the installed SDKs; connecting real Upstash/OpenAI credentials is the only remaining... 30 分钟。

Java 8+ agentic SDK：统一的 LLM 访问（OpenAI/Anthropic/DashScope/Doubao/DeepSeek...），统一的 AI 服务（图像/视频/音频...），Tool Calling、MCP、RAG、SKILLS、Agent Runtime、A2A、agent...

测试发现：All 10 modules build, install, and pass their test suites (1624+ total tests, 0 failures, 10 skipped). The CLI JAR starts and responds to --help and extension list. The docs-site Docusaurus build requires Node.js >=20 (has 18.19.1) which is a preexisting constraint. 19 分钟。

Memento MCP：面向 LLM 的知识图谱记忆系统

测试发现：npm install and TypeScript build succeed; 287/287 unit tests pass with mocked Neo4j; the MCP server binary starts and accepts MCP initialization but cannot connect to Neo4j (not available in this container). 6 分钟。

通过智能语义缓存，将 LLM 成本降低高达 80%，并实现亚毫秒级响应。一个即插即用、与供应商无关的 LLM 代理，使用 Go 编写，响应时间低于毫秒。

测试发现：PromptCache builds, all 42 tests pass across 7 packages, the server starts on port 8080 with health endpoints responding 200, and semantic caching works end-to-end against a mock OpenAI API (cache miss on first request, cache hit on second identical request). 17 分钟。

向量搜索引擎 benchmark 框架

测试发现：All 11 parser tests pass, the run.py CLI shows help, and all 118 engine + 58 dataset configs load. The repo is ready to use with running backend servers. 3 分钟。

即插即用的 LLMs 记忆，仅需 3 行代码。在几分钟内为任何模型添加持久、智能、类人的记忆与回忆。

测试发现：memlayer 0.1.8 installs from source in a venv, all 58+ symbols import cleanly, core storage (ChromaDB + NetworkX) and search (SearchService) work end-to-end, entity deduplication merges 'Dr. Watson' into 'Dr. Emma Watson', SalienceGate correctly classifies salient vs. non-salient text in both LOCAL and LIGHTWEIGHT... 16 分钟。

用于探索、管理和分析向量数据库的现代桌面应用

测试发现：npm ci completes successfully, Vite builds produce valid bundles, the Electron app launches on Xvfb with a visible VectorDBZ 1400x900 window connected to the Vite renderer dev server. The embedding mock server serves embedding vectors on port 3005. Integration tests require Docker containers not available in this... 30 分钟。

Qdrant 向量搜索的 agent 技能：扩展、性能优化、搜索质量、监控、部署、模型迁移、版本升级以及跨 Python、TypeScript、Rust、Go、.NET、Java 的 SDK 使用

测试发现：The repository builds its static skills site into public/ and passes all 16 link-rewriting unit tests with zero validation errors, and its skill-eval harness reports 30/30 (100%) when driven by a mocked claude CLI; the only unverified path is real Claude Code eval scoring. 2 分钟。

基于纯向量搜索的 Graph RAG，在多跳推理场景中实现了 SOTA 性能。

测试发现：The full test suite (143 tests across 11 files) passes cleanly and the FastAPI server starts and responds correctly on /health. 12 分钟。

Vector Storage 是一个向量数据库，能够在浏览器的本地存储中对文本文档进行语义相似度搜索。它使用 OpenAI embeddings 将文档转换为向量，并允许基于余弦相似度搜索相似文档。

测试发现：npm install completes, npx tsc and npm run lint pass, and node verify.mjs exits 0 with ALL CHECKS PASSED covering add/search/filter/hits/persistence/LRU and the OpenAI embedding fetch path against a local mock server. 10 分钟。

K8ssandra 的 Kubernetes operator

测试发现：Go 1.26.10 installed, project builds, 25 pkg packages pass tests, all API validation tests pass, secrets-webhook unit tests pass, yq tests pass. Multi-cluster controller tests (k8ssandra, control, replication) fail intermittently due to envtest resource contention in container environment, they require 3-4 concurrent... 38 分钟。

基于 AI 的个人成长轨迹分析系统，通过记录生活、回顾历史、分析模式帮助用户更好地理解自己，实现持续成长。包含智能路由、RAG 检索、自主任务 agent 等功能。

测试发现：Spring Boot backend compiles and passes 17/18 tests; Vue.js frontend builds; 1 test skipped for PostgreSQL/pgvector dependency. 35 分钟。

一个无代码 RAG 服务器，让答案与你的文档保持同步。支持 8 种向量数据库。基于 Pathway 构建。

测试发现：All 394 non-integration tests pass under the mock license patch, including DuckDB-backed end-to-end indexer tests with persistence, the combined serviette up server+indexer, and the FastAPI server. 86 分钟。

开源 Deep Research 替代方案，用于在私有数据上进行推理和搜索。使用 Python 编写。

测试发现：DeepSearcher package installs with uv sync and pip install -e . and all 422 unit tests pass after fixing 2 test files with stale default model values. 5 分钟。

Pinecone 向量数据库的官方 Python SDK

测试发现：The Pinecone Python SDK builds and imports (v10.0.0), its full unit suite passes (9623 passed, 13 skipped), and the sync and async quickstart document flows plus the gRPC upsert/query path all return correct results against schema-correct local mocks. 36 分钟。

Redis Vector Library。面向 AI 的 Redis Python 原生客户端。

测试发现：redisvl 0.27.2 installed in a venv with all unit tests passing (1619/27skipped), CLI responding correctly, and optional dependencies for MCP, HF, GoogleGenAI, and AzureOpenAI vectorizers resolved. 19 分钟。

使用 LEANN 在一切上运行 RAG。在个人设备上运行快速、准确且 100% 私密的 RAG 应用，同时节省 97% 的存储空间。

测试发现：LEANN Python packages installed, HNSW backend with custom-built Faiss SWIG module works, CI tests pass, README HNSW examples pass with mock embeddings, DiskANN backend blocked by missing libaio system dependency. 87 分钟。

pingcap/autoflow 是一个基于 Graph RAG 的对话式知识库工具，构建于 TiDB Serverless Vector Storage 之上。

测试发现：无法验证；log 显示了它停在哪里。 29 分钟。

AI 时代的 Supabase。一个模块化、开源的 AI 原生软件后端，为知识而非静态数据设计。

测试发现：无法验证；log 显示了它停在哪里。 1 分钟。

自托管 AI 网关，用于私有 RAG、自然语言数据访问和工具调用 agent。

测试发现：无法验证；log 显示了它停在哪里。 28 分钟。