Quivr 2: the RAG project rebuilt as a headless engine for incoming content and alerts
GitHub describes it as Opiniated RAG for integrating GenAI in your apps 🧠 Focus on your product rather than the RAG. Easy integration in existing products with customisation! Any LLM: GPT4, Groq, Llama. Any Vectorstore: PGVector, Faiss. Any Files. Anyway you want.. The repository metadata lists Python as its primary language. The metadata lists the NOASSERTION license. This article stays within the project description and details documented in the GitHub repository README.
At a glance
- What is it?
- QuivrHQ/quivr still carries a description about opinionated RAG with any LLM and any vector store, but Quivr 2 is something else: a Go engine with an HTTP API that stores every version of incoming documents, searches them and fires alerts. It has no user interface, and its plugin model and local requirements shape who can adopt it.
- Who is it for?
- Quivr 2 fits developers building their own application on top of a document store that needs versioning, keyword and semantic search, and topic alerts delivered as signed webhooks, and who are willing to run Docker, PostgreSQL, object storage and plugin services. It does not fit anyone expecting the older drop-in RAG library the repository description still advertises, or a product with its own interface.
- Can I use it commercially?
- Yes. MIT is a permissive licence: you can use, modify and sell software built on it, as long as you keep its copyright and licence notices.
- Is it still maintained?
- Yes. The repository last received commits 6 days ago.
- What is it written in?
- Mainly Python, according to GitHub's language statistics.
Answers come from the project's GitHub data, last synced on November 15, 2026, and from our analysis. They are not legal advice.
Star history
- QuivrHQ/quivr39,651 ★ · +254
| Repository | 2026-08-13 | 2026-11-15 |
|---|---|---|
| QuivrHQ/quivr | 39.4k | 39.7k |
254 Stars in the last 94 days, measured from the daily GitHub snapshot. Compare with other projects
Editorial analysis
The repository description still advertises the old RAG library
The GitHub description reads "Opiniated RAG for integrating GenAI in your apps" and promises any LLM, any vector store such as PGVector or Faiss, and any files. The README for Quivr 2 describes a different product: an open-source engine for content that keeps arriving, which stores documents, makes them searchable and alerts you when something new matches a followed topic. The code has also moved. The clone URL and the Go module path both point to The-Vibe-Company/quivr, the AI agency created by the Quivr team, and the documentation lives at docs.quivr.thevibecompany.co rather than the core.quivr.com homepage. If you arrived looking for a Python RAG helper, Quivr 2 is not that, and older tutorials will not match the current code.
No interface of its own: everything goes through /v0
Quivr has no user interface. Your application talks to it over an HTTP API under /v0, and from 2.0.0 onward that API stays compatible across 2.x releases while upgrades keep your data in place. The OpenAPI contract sits in contracts/, and client/ holds a Go client generated from it. For agents, quivr mcp lets an agent search Quivr and quote passages with their sources.
The consequence is that a team without a front end of its own has nothing to hand to end users. make demo starts a demo web app, quivr-search/, at http://127.0.0.1:5183, but it is a demo built on the public API, not a supported product.
Confirmed means stored, and processing happens afterwards
Quivr confirms a document only after it has been written to PostgreSQL and object storage, then processes it in the background through the Quivr worker. An outage delays that processing but does not lose anything already confirmed, and the search index can always be rebuilt from storage. Every version is kept: a correction becomes a new version and the old one stays readable, while a withdrawn document leaves search immediately and any alerts that caught it are notified. The trade-off is timing. A successful API response means the document is safe, not that it is already searchable, so an application that writes and then instantly searches should expect a short gap.
Alerts take boolean queries, plain language and signed webhooks
Following a topic is the feature that separates Quivr 2 from a plain search store. You save an alert either as a query such as "harbour strike" AND (ferry OR port) or as a plain-language description, and Quivr records every new document that matches along with the reason. Your application either reads those matches or receives them as signed webhooks. Search itself runs by keywords, by meaning, or both, and every result points to the exact passage, document and version, which keeps citations traceable. Content arrives one document at a time, in batches, or from scheduled sources: RSS feeds, Microsoft 365 mailboxes and X lists.
Plugins are separate HTTP services you have to run
File formats, sources, embedding models, ranking and alert rules are all plugins, and a plugin is a small HTTP service rather than code loaded into the engine. That makes them replaceable and writable in any language, with Go and Python kits in sdks/ and first-party plugins for alerts, sources, PDF, ingestion and retrieval in plugins/. Even reading a PDF goes through a plugin. The cost is operational: each plugin is another service to deploy, monitor and upgrade alongside the API, worker, PostgreSQL and object storage. A small team that wanted a single process will be running several.
The local stack needs Linux x86_64 or Apple Silicon and about 1 GB
Trying Quivr from a checkout requires Linux on x86_64 or a Mac with Apple Silicon, Docker with Compose v2, the Go version in go.mod (go 1.27.3), Python 3.12 with venv, and jq. Windows and Intel Macs are not on that list, so contributors on those machines need another host. The first start downloads about 1 GB:
git clone https://github.com/The-Vibe-Company/quivr.git && cd quivr
make dev # start Quivr and its dependencies in Docker
eval "$(make -s env)" # point this shell at the API, with a local keyThe Quickstart then walks through a collection, an article, keyword and semantic search and an alert in about fifteen minutes. Server installs from published images use the Run Quivr on one machine guide instead of a checkout.
Three releases in two days around 2.0.0
Quivr 2.0.0 shipped on 2026-10-29, followed by v2.0.1 and v2.0.2 on 2026-10-30, and the main branch was last pushed on 2026-11-02. Releases are driven by release-please, with version.txt and CHANGELOG.md at the root. Early patch releases straight after a major version are a reason to pin an exact 2.0.x image and read the changelog before upgrading. For contributors, make check runs everything that needs no Docker stack in about two minutes, make verify runs the end-to-end journeys on an isolated stack, and pull request titles follow Commitizen, for example feat(ingestion): accept record versions. Security issues go through SECURITY.md privately.
Editorial conclusion
Quivr 2 fits developers building their own application on top of a document store that needs versioning, keyword and semantic search, and topic alerts delivered as signed webhooks, and who are willing to run Docker, PostgreSQL, object storage and plugin services. It does not fit anyone expecting the older drop-in RAG library the repository description still advertises, or a product with its own interface. Before adopting it, read the Known limits and Versions and compatibility pages on docs.quivr.thevibecompany.co and confirm your machines match the supported platforms.
Frequently asked questions
What is Quivr used for?
Quivr 2 stores incoming documents such as news articles, emails, posts and PDFs, keeps every version, makes them searchable by keyword and meaning, and alerts your application when new content matches a saved topic.
Is Quivr free to use?
Quivr is open source under the MIT license. You run it yourself with Docker, PostgreSQL and object storage, either from a checkout or from published images.
What is Quivr 2 built from?
Quivr 2 is a single Go binary, cmd/quivr, that runs the API, worker, migrations, plugin tools, search and an MCP server. Documents go into PostgreSQL and object storage, and plugins run as separate HTTP services.
Official sources
Where this project is recommended
Add this badge to your README
If you maintain this project, the badge below links readers to this analysis and shows its maintenance status from the daily GitHub snapshot. Paste the markdown into your README; add ?metric=license or ?metric=stars to the image URL for a different field.
[](https://hysenlabs.com/projects/quivrhq-quivr)Star history chart for your README
[](https://hysenlabs.com/projects/quivrhq-quivr)Was this useful?