Key Takeaways
- Apache 2.0 license — the source code (github.com/QuivrHQ/quivr, now hosted at github.com/The-Vibe-Company/quivr) is free to use, modify, and fork
- No commits to the GitHub main branch since June 19, 2025 — no formal shutdown announcement exists, but there has been no visible development for over a year
- The GitHub organization was renamed from QuivrHQ to The-Vibe-Company; old QuivrHQ links redirect automatically
- quivr.com, the company website, now markets an unrelated AI customer-support agent product with no mention of the open-source RAG library
- pip install quivr-core still works and installs version 0.0.26; the code runs, but no one is fixing bugs or reviewing pull requests
- For a turnkey self-hosted chat interface, evaluate AnythingLLM or Khoj instead — both ship a full UI and are actively committed to as of this review
📍 In One Sentence
Quivr is an open-source, Apache 2.0 Python library (quivr-core) for building RAG into your own app, not a self-hosted chat interface, and its GitHub main branch has had no commits since June 19, 2025.
💬 In Plain Terms
Older write-ups describe Quivr as a browser-based "second brain" you install and chat with — that version is gone from the repository. What ships today is a Python package you import into your own code.
⚠️Warning: If you are looking for a self-hosted "chat with your files" web app, Quivr in its current form is not that. It is a Python library with no bundled frontend.
Is Quivr Still Maintained?
No formal announcement of a shutdown or archival exists for Quivr, unlike some other open-source AI tools that have published a dated wind-down notice. But the observable evidence points the same direction: the GitHub main branch has had no commits since June 19, 2025, and the company behind the project now markets an unrelated product.
The repository itself is not marked "archived" on GitHub — it still accepts issues and discussions, and the Apache 2.0 code is fully installable today. Treat this as a dormant, unmaintained project rather than a formally discontinued one: the distinction matters for how much risk you accept by depending on it.
Last commit to main branch
- Finding:
- June 19, 2025 — no commits since, per the GitHub commits API
GitHub organization
- Finding:
- Repository moved from QuivrHQ to The-Vibe-Company; old links auto-redirect
Latest PyPI release
- Finding:
- quivr-core 0.0.26, published December 10, 2024
Untagged newer version
- Finding:
- core-0.0.33 was tagged on GitHub February 4, 2025 but never published to PyPI
Repository status
- Finding:
- Not formally archived — issues (35 open) and discussions remain enabled
Company's current product
- Finding:
- quivr.com now markets an unrelated AI customer-support agent product, with no reference to the RAG library
📌Note: This changes what you should expect from every section below: the Apache 2.0 code runs today, but treat it as a frozen dependency with no active support, not as an actively developed product.
What Is Quivr?
Quivr is an open-source Python library (Apache 2.0 license, published as quivr-core on PyPI) for adding retrieval-augmented generation to your own application through code — not a standalone web app you deploy and open in a browser. The project describes itself as helping developers build a "second brain," but that second brain is a library you import, not a hosted product.
- Installed with pip install quivr-core — no Docker, frontend, or backend service ships in the current repository
- Central abstraction is the Brain class: Brain.from_files(...) ingests documents, brain.ask(...) queries them
- Works with OpenAI, Anthropic, and Mistral APIs directly, or with local models through Ollama, per the project README
- Ingests PDF, TXT, and Markdown natively, and can be extended with custom parsers or the companion Megaparse project for harder documents
- Retrieval workflows (history filtering, query rewriting, retrieval, generation) are configured through a YAML file, not a settings screen
- Optional reranker step (Cohere is the documented integration) to reorder retrieved chunks before generation
What Document Types, LLMs, and Vector Stores Does Quivr Support?
Quivr accepts PDF, TXT, and Markdown files out of the box, connects to OpenAI, Anthropic, and Mistral APIs or a local Ollama model, and integrates with PGVector and FAISS for vector storage, per the project's own documentation and repository description.
- Documents: PDF, TXT, Markdown natively; other formats through a custom parser you write, or through Megaparse (a separate QuivrHQ project) for OCR-heavy or complex layouts
- LLM APIs: OpenAI, Anthropic, and Mistral are named explicitly in the README as supported providers
- Local models: Ollama is documented as a supported backend, letting you run the LLM step without a cloud API call
- Vector stores: the repository description names PGVector and FAISS; the project is designed to be vectorstore-agnostic, but wiring an alternative is a configuration task you do yourself
- Reranking: an optional Cohere reranker step can reorder retrieved chunks before they reach the LLM
How Do You Install and Self-Host Quivr Now?
Quivr has no server to deploy — "self-hosting" means running the Python library inside your own script, service, or application, entirely on infrastructure you control. These steps get a working Brain answering questions from local files.
- 1Install Python 3.10 or newer, then install the package: pip install quivr-core.
- 2Set an API key as an environment variable for whichever LLM provider you plan to use — OPENAI_API_KEY, ANTHROPIC_API_KEY, or a Mistral key — or point Quivr at a local Ollama endpoint to avoid a cloud LLM call entirely.
- 3Create a Brain from your files: Brain.from_files(name="my_brain", file_paths=["doc1.pdf", "doc2.md"]) ingests and indexes the documents in-process.
- 4Write a workflow_config.yaml file defining the retrieval steps (filter_history, rewrite, retrieve, generate) plus optional reranker and LLM settings such as max_input_tokens and temperature.
- 5Load the config with RetrievalConfig.from_yaml("workflow_config.yaml") and call brain.ask("your question", retrieval_config=retrieval_config) to query your documents.
- 6Persist the index yourself if you need it to survive a restart — quivr-core keeps everything in memory for the life of the script unless you wire up a vector store like PGVector or FAISS for storage.
Do I need a server or Docker to run Quivr?
No. quivr-core is a Python package you import into a script or application — pip install quivr-core is the entire deployment. There is no bundled frontend, backend service, or docker-compose file for a browser-based chat interface, which is a meaningful difference from AnythingLLM or Khoj.
Can Quivr run fully offline with a local LLM?
Yes, for the generation step — the README documents Ollama support for local models, so the LLM call itself can stay on your machine. The optional Cohere reranker step is a cloud API call unless you skip it.
Who Should Use Quivr?
The current shape of the project — a code-first library with no active maintenance — narrows who Quivr fits. It is a reasonable pick for one specific kind of user, and a poor one for most people searching for a self-hosted document-chat app.
Quivr vs. Alternatives
Every alternative below ships a deployable chat interface and had commits within the past month of this review — the opposite of Quivr's current state.
Tool | Interface | License | Best For | Maintenance |
|---|---|---|---|---|
| Quivr | Python library, no UI | Apache 2.0 | Devs embedding RAG in their own app | No main-branch commits since mid-2025 |
| AnythingLLM | Full chat UI, Docker | MIT | Turnkey self-hosted knowledge assistant | Active |
| Khoj | Chat UI + agents | AGPL-3.0 | Self-hosted "second brain" with automations | Active |
| PrivateGPT | API + chat UI | Apache 2.0 | Strict offline / air-gapped deployments | Active |
Common Mistakes When Evaluating Quivr Now
These mistakes come from treating older write-ups about Quivr — including its own historical marketing — as a description of what the repository ships today.
Frequently Asked Questions
Is Quivr still maintained?
There is no formal shutdown announcement, but the GitHub main branch has had no commits since June 19, 2025, and the company behind it (now trading as The Vibe Company) markets an unrelated AI customer-support product on quivr.com as of this review. Treat it as dormant, not actively developed.
Is Quivr a self-hosted app or a library?
A library. quivr-core is a Python package (pip install quivr-core) built around a Brain class for retrieval-augmented generation. The current repository ships no frontend, backend service, or docker-compose file.
What license is Quivr released under?
Apache License 2.0, per the LICENSE file in the repository. GitHub's automatic license badge shows "Other" for this repository, but the LICENSE file text itself is the standard Apache 2.0 license.
Does Quivr support local LLMs?
Yes. The README documents Ollama support for local models, alongside direct API support for OpenAI, Anthropic, and Mistral.
What file types can Quivr ingest?
PDF, TXT, and Markdown natively. Other formats require a custom parser you write, or the companion Megaparse project for harder documents like scanned PDFs.
What vector databases does Quivr support?
The repository description names PGVector and FAISS. The project is designed to be vectorstore-agnostic in principle, but wiring up an alternative store is a configuration task left to you.
How is Quivr different from AnythingLLM or Khoj?
AnythingLLM and Khoj both ship a deployable chat interface and had commits within the past month of this review. Quivr ships only the Python library, with no bundled frontend and no commits to its main branch since June 2025.
Can I still install Quivr today?
Yes. pip install quivr-core installs version 0.0.26, and the code runs as documented. You will not get bug fixes, new integrations, or support going forward.
Is quivr.com the same as the open-source project?
It is run by the same company. As of this review, quivr.com markets an AI customer-support agent product and makes no mention of the open-source RAG library or a self-hosted knowledge assistant.