swirl-search

swirlaiApache-2.0活跃维护

不用搬数据就能做企业知识检索和 RAG,跨 100+ 应用联合搜索并直接生成答案,几分钟即可部署。

AI Search & RAG Without Moving Your Data. Get instant answers from your company's knowledge across 100+ apps while keeping data secure. Deploy in minutes, not months.

v4.5.0.7发布于 2026-06-22 15:32 UTC3,049star291fork同步于 2026-11-11 05:40 UTC

怎么上手

由 README 提炼,执行前请确认
curl https://raw.githubusercontent.com/swirlai/swirl-search/main/docker-compose.yaml -o docker-compose.yaml
docker-compose pull && docker-compose up
硬件要求按 README 说明

README 未说明显存或 GPU 要求;官方推荐用 Docker 部署,未提及 CPU 运行细节。

下载:源码 zip·GitHub Releases(安装包)本站不镜像安装包

项目说明

由 README 自动翻译提炼

不搬数据的联邦 AI 搜索与 RAG,实时查询 100+ 应用并给出带来源的答案

不用搬数据就能做企业知识检索和 RAG,跨 100+ 应用联合搜索并直接生成答案,几分钟即可部署。

核心能力

  • 实时联邦查询 100+ 连接器,数据留在原处
  • 用自带 OpenAI Key 生成带引用的 RAG 答案
  • 用 spaCy 与 NLTK 的余弦相似度做重排序和去重
  • 通过 Processor 流水线改写查询、响应与结果
  • 提供同步或异步联邦的 REST API,结果存入 SQLite 或 Postgres

容易踩的坑

Docker 版关闭后不保留数据和配置,需要持久化要改用 Quick Start Guide 的安装方式;启用实时 RAG 需自行设置 OPENAI_API_KEY 等环境变量;默认登录为 admin/password,服务开在 8000 端口。

内容由模型根据 README 生成,可能不完整或不准确,以官方文档为准。看原文 README

最近发版

5 个版本v4.5.0.7
  1. v4.5.0.7SWIRL Community 4.5.0.7最新正式版2026-06-22 · 2026-06-22 15:32 UTC

    SWIRL Community 4.5.0.7 — Security Patch Fixes #1941:

    • OIDC privilege escalation: sign-in no longer provisions superuser/staff accounts, and no longer sets a shared hardcoded password.
    • Cross-user Microsoft token fallback: M365 token lookup is now strictly scoped to the requesting user; removed the fallback that could use another user's credentials.

    No schema or dependency changes. Upgrade recommended for any deployment using OIDC sign-in or M365 connectors.

  2. v4.5.0.6SWIRL Community 4.5.0.62026-05-22 · 2026-05-22 15:34 UTC
    查看更新说明

    SWIRL Community 4.5.0.6

    Fixes the AI Summary 'appears then swaps to a different summary' bug surviving 4.5.0.5.

    • Galaxy: AI Summary toggle no longer triggers a duplicate search.
    • Backend: per-search_id lock guarantees one RAG completion per search, even under concurrent requests.

    ⚠️ No DB migration. Drop-in for any 4.5.0.x.

    docker pull swirlai/swirl-search:latest
    
  3. v4.5.0.5SWIRL Community 4.5.0.52026-05-21 · 2026-05-21 02:28 UTC
    查看更新说明

    SWIRL Community 4.5.0.5 — RAG cache regression fix

    Follow-up patch on top of 4.5.0.4 that fixes a long-standing RAG UX bug exposed by 4.5.0.4 smoke testing: intermittent: the AI Summary appears, then a few seconds later gets replaced by a different AI Summary.

    ⭐ PLEASE STAR OUR REPO: https://github.com/swirlai/swirl-search 🌐 PLEASE VISIT OUR WEBSITE: https://www.swirlaiconnect.com/


    What's fixed

    Backend — swirl/views_helpers/search_rag.py (SearchRag.get_rag_result()):

    • The cache check now correctly treats an empty rag_query_items (the typical state of Galaxy's follow-up GET /sapi/detail-search-rag/?search_id=…) as 'no filter requested, serve whatever's stored' rather than 'items changed, regenerate'.

    Upgrading

    ⚠️ No database migration required. 4.5.0.5 is a drop-in replacement for 4.5.0.4 (and for any earlier 4.5.0.x).

    Docker users:

    docker pull swirlai/swirl-search:latest
    

    See the Docker Build Guide for the SBOM / vulnerability inspection commands.

  4. v4.5.0.4SWIRL Community 4.5.0.42026-05-20 · 2026-05-20 22:37 UTC
    查看更新说明

    SWIRL Community 4.5.0.4 — Auth-chain regression fix

    This release resolves an issue introduced in SWIRL 4.5.0.3.

    ⭐ PLEASE STAR OUR REPO: https://github.com/swirlai/swirl-search 🌐 PLEASE VISIT OUR WEBSITE: https://www.swirlaiconnect.com/


    Fixed

    • SWIRL now quietly ignores stale credentials your browser may be carrying and just uses your active login session instead of throwing you out.

    Upgrading

    ⚠️ No database migration required. 4.5.0.4 is a drop-in replacement for 4.5.0.3 (and for 4.5.0.2 if you skipped 4.5.0.3).

    Docker users:

    docker pull swirlai/swirl-search:latest
    

    See the Docker Build Guide for the SBOM / vulnerability inspection commands.

  5. v4.5.0.2SWIRL Community 4.5.0.22026-05-20 · 2026-05-20 04:45 UTC
    查看更新说明

    SWIRL Community 4.5.0.2

    Targeted patch that adds the optional AI instruction field on the search form, resolves minor RAG/UX issues in the bundled Galaxy UI, and pins major dependencies to prevent surprise version jumps.

    ⭐ PLEASE STAR OUR REPO: https://github.com/swirlai/swirl-search 🌐 PLEASE VISIT OUR WEBSITE: https://www.swirlaiconnect.com/


    New Features

    This version of SWIRL offers a separate field for entering instructions to the LLM.

    When you click the Generate button, it will provide those instructions along with the most relevant search results to the LLM, thus separating the retrieval and instructions, and guiding the response:

    All responses will include citations. Click them to verify the results.

    To ask a follow up question, change the question in the instruction box, and click Generate again.

    Improvements

    • History toggle moved to the left of the SWIRL logo
    • Login page logo refreshed to the canonical Galaxy logo
    • Key package versions now pinned in requirements.txt

    Bill of Materials (BOM)

    To obtain the Bill of Materials for SWIRL Community:

    docker scout sbom swirlai/swirl-search:latest
    

    Upgrading

    ⚠️ No database migration required. 4.5.0.2 is a drop-in replacement for 4.5.0.1.

用户点评

登录用户的真实使用记录,可按版本和硬件筛选
共 0 条

    还没有人点评。第一个说出真实体验的人最有价值。

    README 原文

    未翻译

    SWIRL

    SWIRL Community

    Federated AI search and RAG across your stack, without moving your data

    Ask a question. SWIRL searches your apps live, ranks the results, and returns an answer with sources you can click through to. Data stays where it lives. No vector database, no ETL, no second copy to govern. Open source, Apache 2.0, running on your machine in about two minutes.

    License: Apache 2.0 GitHub Release GitHub Stars Website

    SWIRL Galaxy UI: ask a question, get an answer with sources

    ⚡ Quick Start · 🧱 What you get · 🆚 Community vs Enterprise · 🔌 Connectors · 🤝 Contribute

    This repository is SWIRL Community, Apache 2.0 and free to self-host. There is also SWIRL Enterprise, which adds a three-pass reranker, canonical answers, an MCP server for agents, and managed support. The comparison below lays out the difference, so you can tell which one you need.

    🤔 Why SWIRL?

    Most "AI search" asks you to copy everything into a vector database first, then govern that copy forever. SWIRL skips the copy. It queries your sources live, with the user's own permissions, re-ranks the results, and optionally generates an answer with citations using the LLM of your choice.

    ❌ The usual way✅ With SWIRL
    Stand up a vector databaseNo vector DB needed
    Move and duplicate your dataData stays in place
    Build ETL pipelinesQuery live, in place
    Weeks of infrastructure workOne Docker command, about 2 minutes
    A new copy to secure and auditPermissions enforced at the source

    🔥 Quick Start: Docker in 2 minutes

    Make sure the Docker app is installed and running.

    Download the compose file:

    curl https://raw.githubusercontent.com/swirlai/swirl-search/main/docker-compose.yaml -o docker-compose.yaml
    

    Optional, to enable real-time RAG with your own OpenAI key:

    export MSAL_CB_PORT=8000
    export MSAL_HOST=localhost
    export OPENAI_API_KEY='<your-OpenAI-API-key>'
    

    Start SWIRL:

    docker-compose pull && docker-compose up
    

    Then open http://localhost:8000/galaxy, log in with admin / password, and run a search. SWIRL comes ready to search Arxiv, European PMC and Google News out of the box.

    SWIRL Galaxy UI: federated, ranked results across your sources

    Note: the Docker version does not retain data or configuration when shut down. For a persistent install, see the Quick Start Guide. Watch the 60-second video tutorial to get going.

    🧱 What SWIRL Community gives you

    Apache 2.0, free, yours to run anywhere:

    • Federated search and RAG across 100+ connectors, with your data left in place.
    • The Galaxy UI, the same interface that powers SWIRL Enterprise.
    • Real-time RAG with citations using your own OpenAI key.
    • Microsoft 365 integration with OAuth2.
    • Re-ranking with cosine vector similarity (spaCy large model plus NLTK), duplicate detection, and result mixers.
    • A pipelined Processor architecture for transforming queries, responses and results.
    • Synchronous or asynchronous federation over a clean REST API.
    • Results stored in SQLite or Postgres for post-processing and analytics.
    • Easily extensible Connector and Mixer objects, so adding a source is straightforward.

    What teams build with it

    • Knowledge base search across SharePoint, Confluence and Drive, with source links.
    • Customer support assistants that search docs and tickets and draft grounded responses.
    • Developer assistants over GitHub, Jira and documentation.
    • Unified search across every tool, with results that respect existing permissions.

    🆚 Community vs SWIRL 5 Enterprise

    Community is genuinely useful and genuinely open. SWIRL 5 Enterprise is what you graduate to when search becomes infrastructure. We keep the line honest so you always know what you are running.

    CapabilityCommunity (Apache 2.0)SWIRL 5 Enterprise
    Federated search and RAG, no data movementYesYes
    Galaxy UIYesYes
    100+ connectorsYesYes, plus managed connectors
    RAG with your own LLM keyYesYes
    Relevancy rankingCosine similarity (spaCy, NLTK)Three-pass pipeline: BM25, then E5 embeddings with hybrid fusion, then a cross-encoder
    Canonical answers and Pinned ResultsNot includedYes
    First-class MCP server for agentsNot includedYes
    Hallucination warning on generated answersNot includedYes
    Business console with AI-Yield analytics, semantic cache and dedup at scaleNot includedYes
    SOC-2 hosting, managed connectors, and supportNot includedYes

    If you outgrow cosine ranking, want canonical answers, or need your agents to call SWIRL over MCP, that is SWIRL 5.

    When you outgrow Community

    Plenty of teams run Community in production and never need more. If you do reach its edges, where you want the three-pass reranker, canonical answers, an MCP server for your agents, the hallucination warning, or managed connectors and support, that is what SWIRL Enterprise adds.

    If that is where you are heading, you can talk to us for a walkthrough on your own systems. Nothing leaves your environment, and there is no obligation.

    🔌 Connectors

    The full, current list lives at swirlaiconnect.com/connectors. Connectors are easily extensible; see the Connector objects and the Developer Guide. To request a connector, email [email protected].

    📖 Documentation

    Full docs: docs.swirlaiconnect.com. Start with the Quick Start Guide, then the User Guide and Developer Guide.

    🤝 Contributing

    SWIRL is built in the open and we welcome contributions. Good places to start:

    • Browse open issues and look for good first issue.
    • Join GitHub Discussions to ask questions and share what you are building.
    • Write a connector or a mixer; the objects are designed to be extended.

    Please read the Code of Conduct and Contributing guide in the repo before opening a pull request.

    👷 Support and community

    License

    SWIRL Community is licensed under the Apache License 2.0. See LICENSE and NOTICE for details.