Skip to main content
PromptQuorumBuilt for humans. Structured for AI.
Home/Power Local LLM
RAG & Document Chat

OpenRAG Review: Open-Source Agentic RAG Platform

·7 min readBy Hans Kuepper · Founder of PromptQuorum · Discovery engine for open-weight & open-source AI

OpenRAG (version 0.8.0, released 1 October 2026) is a free, Apache-2.0 platform that bundles Langflow, OpenSearch and Docling so you can upload documents and chat with them through a web interface. It runs as containers started from a terminal, which makes it a developer tool rather than a double-click app. PromptQuorum has not installed or tested it; every statement here comes from the project's own documentation.

Key Takeaways

  • Maintenance: v0.8.0 was published on 1 October 2026 and the repository changed again on 23 October 2026, so no stale-project warning applies.
  • License: Apache-2.0, with the source on GitHub under the langflow-ai organization.
  • Cost: the package has no price; your costs are hardware and any model API you connect.
  • Windows: supported only inside the Windows Subsystem for Linux.
  • Weight: it is container-based and disk-hungry, so it suits a dev machine or server more than a light laptop.

📍 In One Sentence

OpenRAG is an Apache-2.0 package that starts a ready-made RAG stack for your documents with one uvx command.

💬 In Plain Terms

You run a terminal installer, it launches a few containers, and you then upload files and ask questions in a browser chat.

📌Note: This review is based on the OpenRAG documentation, website and GitHub release page, checked on 23 October 2026. PromptQuorum has not installed, tested or benchmarked the software.

Read in:
🇺🇸en🇩🇪de🇫🇷fr🇯🇵ja🇨🇳zh🇪🇸es🇧🇷pt🇸🇦ar🇰🇷ko

OpenRAG is an open-source package that wires Langflow, OpenSearch and Docling together so you can upload documents, search them and chat with them in a browser. It is started from a terminal and runs as containers, so it is aimed at developers. This review reflects its documentation, website and GitHub release page as checked on 23 October 2026.

OpenRAG Review: Open-Source Agentic RAG Platform

What Is OpenRAG?

OpenRAG is a retrieval-augmented generation platform in which you add documents to a knowledge base and ask an AI agent questions about them in a chat interface. The project describes it as an open-source package for building agentic RAG systems that combines three existing projects: Langflow for agent flows, OpenSearch for document and vector storage, and Docling for parsing and chunking files.

It is developed in the langflow-ai/openrag repository, with a FastAPI backend and a Next.js frontend. Its documentation lives at docs.openr.ag.

Get It

OpenRAG is not a downloadable desktop app: you install it from a terminal with uv, and it then starts its own containers.

Option
How to get it
Quickstart (macOS, Linux)Run uvx --python 3.13 openrag in an empty folder
WindowsSame command, but only inside WSL (see Requirements)
Python projectAdd it as a dependency with uv, per the install options
Self-managedDocker or Podman with your own .env file, per the Docker guide
Websiteopenr.ag
Documentationdocs.openr.ag
Source codegithub.com/langflow-ai/openrag
Privacy policyNo own page found; the docs cite IBM's privacy statement

Version reviewed: v0.8.0, published on 1 October 2026 per the GitHub release page; the repository itself was last updated on 23 October 2026. The release ships only a Python wheel and source archive, with no installer or desktop build, so a terminal and a developer environment are required.

Requirements

The docs ask for a container runtime, 8 GB of RAM and more than 50 GB of free disk space.

Item
What the docs say
Python and uvPython 3.13 and uv, per the quickstart
Container runtimeDocker or Podman plus a compose tool; uvx installs one if missing
RAM8 GB is the minimum recommended
DiskMore than 50 GB free for onboarding and ingestion
WindowsInside WSL only; not on a WSL distro in a Windows VM
Local modelsgpt-oss:20b via Ollama needs at least 16 GB of RAM

The pages checked state no GPU requirement for OpenRAG itself, only that Ollama must run in a GPU-enabled container to use a GPU.

How to Get Started

The steps follow the project's quickstart; PromptQuorum has not run them.

  1. 1
    Create a workspace folder
    Why it matters: Make an empty directory such as openrag-workspace and open a terminal in it.
  2. 2
    Start the installer
    Why it matters: Run uvx --python 3.13 openrag, which opens a terminal interface.
  3. 3
    Set the admin passwords
    Why it matters: Autogenerate or enter the OpenSearch and Langflow admin passwords; skip the other prompts to keep the defaults.
  4. 4
    Start the services
    Why it matters: Answer Y to "Start services now?" and wait for the "Services are running" message, then open OpenRAG in the browser.
  5. 5
    Complete onboarding
    Why it matters: Choose a model provider, enter its API key and click Complete; the quickstart uses OpenAI.
  6. 6
    Add knowledge and chat
    Why it matters: Upload files or folders, then ask questions in the chat view.

Features Confirmed by the Sources

Every item below comes from the project's documentation or README.

  • Editable flows. Built-in agent flows can be customised in the embedded Langflow editor, including swapping models or components, and custom Langflow components and MCP servers are supported.
  • Document ingestion. Files come from your local filesystem or from OAuth-authorised cloud storage connectors; Docling Serve parses and chunks them.
  • Search. OpenSearch stores documents and embeddings and adds hybrid search and multi-tenancy.
  • SDKs and API. Official Python and TypeScript SDKs are published as openrag-sdk, and the docs include an API reference.
  • Observability. Tracing to Langfuse and the IBM Instana tracer are optional and switched off by default.

Data Flow and Access

Per the docs, Docling Serve processes your documents locally and OpenSearch stores them in your own knowledge base; cloud sources load only through connectors you authorise. The architecture is containers: a backend, Langflow, Docling Serve, the connectors and a frontend.

OpenAI is the default model provider, so with it your prompts and the retrieved document excerpts go to OpenAI under its own policy. The configuration reference also lists Anthropic, Ollama and IBM watsonx.ai, which lets you route to a local Ollama endpoint instead. If no Langflow superuser password is set, Langflow starts without authentication, which is why the docs recommend setting one.

📌Note: The local-processing statement is the project's own description; PromptQuorum has not audited the containers or their network traffic. The terminal's upgrade check queries the Docker Hub API unless the image registry is changed.

Who Should Use It

  • Developers who want a pre-wired RAG stack instead of assembling Langflow, OpenSearch and Docling themselves.
  • Teams that already run containers and want hybrid search with multi-tenancy over their own files.
  • Builders who need an API or SDK to call a document-search agent from their own application.

What We Could Not Verify

  • Hands-on quality. PromptQuorum did not run the software, so answer quality, citation accuracy, ingestion speed and stability are not assessed.
  • Telemetry. The pages checked document no analytics or opt-out setting for the containers.
  • Privacy policy. No OpenRAG-specific policy page was found; the docs page only references IBM's general privacy statement.
  • Local-only operation. The docs describe Ollama support, but the quickstart was only documented with OpenAI, so a fully offline setup is untested.

Competitors and Alternatives

Langflow

Platforms:
Web
Price / license:
Free / MIT
Key difference:
The visual flow builder that OpenRAG embeds, without the bundled search stack
Articles about Langflow (3)

Also mentioned in:

RAGFlow

Platforms:
Linux, web
Price / license:
Free / Apache 2.0
Key difference:
RAG engine centred on deep document understanding and citations
Articles about RAGFlow (5)

Also mentioned in:

Competitor details change often; confirm each app's current price, license and platforms on its own listing.

Compare in directory

Frequently Asked Questions

Which Ollama models does OpenRAG recommend?

The docs suggest gpt-oss:20b or mistral-nemo:12b for language and nomic-embed-text, mxbai-embed-large or embeddinggemma for embeddings. Chat models must support tool calling, and the docs do not guarantee every Ollama model works.

Which ports does it use?

By default 3000 for the frontend, 7860 for Langflow, 8000 for the backend, 9200 for OpenSearch, 5601 for OpenSearch Dashboards and 5001 for Docling. Frontend and Langflow ports can be changed in the .env file.

Which cloud storage can it ingest from?

The configuration reference describes Google Drive through Google OAuth, SharePoint and OneDrive through a Microsoft Graph app registration, and AWS S3 through access keys.

How do I call it from my own code?

Install the Python SDK with pip install openrag-sdk or the TypeScript SDK with npm install openrag-sdk. The older standalone openrag-mcp package is deprecated.

Verdict

OpenRAG is a free, Apache-2.0 RAG stack that, per its documentation, ships Langflow, OpenSearch and Docling already connected and is actively updated, with a release in early October 2026 and commits through 23 October. Against that, it is heavy to run, supports Windows only through WSL, defaults to OpenAI, and PromptQuorum could not verify its quality, telemetry or fully offline use. It suits developers who want a ready-made, hackable RAG back end; readers who prefer an installable app can compare AnythingLLM, and those who want only the flow builder can look at Langflow.

Sources

← Back to Power Local LLM

MCP