Humata
ProductivitySearch, summarize and compare uploaded documents with source-linked answers. Humata adds team-oriented document access and page-based plans for ongoing knowledge work.
Questions across filesCited passages
DOCUMENT AI / PARSE · RETRIEVE · ANSWER
RAGFlow
A free self-hosted RAG platform that parses documents, makes retrieval inspectable and connects source-backed information to AI chat and applications.
Official RAGFlow quick-start example of document-based chat with source citationsExplore the image ↓
What you can do with it
What this tool is for
RAGFlow is designed for workflows where finding the right material inside documents matters as much as generating an answer. It parses files into searchable pieces, lets you inspect those pieces and connects relevant results to a language model. This makes it useful for manuals, reports, slides and other collections whose structure can be lost in a simple copy-and-paste chat.
Its role is retrieval-augmented generation, often shortened to RAG. The model receives selected document context when responding; it is not permanently retrained on every uploaded file. Source references improve traceability but do not guarantee correctness. An answer can still omit a condition or misread a table, so the underlying passage remains important.
01
Technical documentation search
02
Report comparison
03
Internal procedure lookup
04
Document ingestion diagnosis
A closer look at the product
Official product image. Click to inspect the details.
Product interface and examples
Key features, practical tasks
Capabilities in practice
- Document parsing: Process supported documents, spreadsheets, images and presentations. Choose a parsing and chunking approach that suits the material; a slide deck and a long prose manual do not necessarily benefit from identical treatment.
- Visible chunks: Inspect the pieces produced from a file before using them in chat. Poor extraction, broken tables or missing headings are easier to fix at this stage than after many confident but incomplete answers.
- Manual retrieval intervention: The guide describes editing chunk text and adding keywords or questions. These adjustments can improve retrieval for important terminology, but should preserve what the original source actually says.
- Retrieval testing: Ask a diagnostic question and examine the returned passages independently of the language model. This separates a search problem from a generation problem and gives a concrete place to improve the application.
- Dataset-specific assistants: Associate a chat assistant with one or more datasets. Use clear dataset names and versions so the assistant does not mix instructions from incompatible product generations.
- Multiple model roles: Configure chat, embedding and other required model types separately. A good answer model does not replace a missing embedding service or correct a poor parsing configuration.
- API integration: Use the documented HTTP API to integrate datasets and conversations into another application. Keep application keys and access boundaries distinct from publicly shared chat content.
Get started, step by step
- 01
Deploy a compatible release
- 02
Select model roles and create a dataset
- 03
Parse and inspect one representative file
- 04
Create a bounded chat assistant
A practical first workflow
A practical first workflow
- Deploy a compatible release. Follow the guide for the chosen release and supported platform. Prepare persistent data and sufficient memory, then confirm the service and its dependencies are healthy.
- Select model roles and create a dataset. Configure an embedding model and a chat model. Choose the dataset’s embedding and chunking approach before indexing substantial material.
- Parse and inspect one representative file. Check extracted chunks, important tables and headings. Try a retrieval question and make focused corrections where the intended passage is missing.
- Create a bounded chat assistant. Choose its datasets, answer model and missing-answer response. Ask both an answerable question and one outside the collection before sharing the application.
Example prompt or task: Using the selected manual, state the supported operating conditions and the exceptions. Cite the corresponding passage for each point. If the retrieved material does not answer the question, say the answer was not found in this dataset.
This editorial task checks retrieval boundaries. A missing-answer setting and careful source review are more useful than assuming the presence of citations makes hallucinations impossible.
Find the right plan for you
Free self-hosting
Explore official plans ↗Free software Compute and model services are separate
The platform is free to deploy under its license. Parsing, storage and model inference still use resources.
What is free and what can cost extra
RAGFlow’s repository uses Apache-2.0, providing a free self-hosted route. The server, database or document engine, stored files, parsing jobs and model requests still need resources. A local inference service can avoid external API billing, but requires additional capacity.
The current quick-start guide recommends a starting point of four x86-64 CPU cores, 16 GB RAM and 50 GB free storage. Treat that as an application starting point, not a guarantee for every workload. Local models, large document collections and concurrency can require substantially more.
A few things worth knowing
Frequently asked questions
Is RAGFlow free?
The Apache-2.0 project can be self-hosted without a software subscription. Infrastructure and model services are separate costs.
Does uploading files train the model?
No. The normal workflow retrieves relevant chunks and places them in the model’s context. That is different from permanently changing its weights.
Why does it cite the wrong passage?
Inspect parsing, chunking, selected datasets and retrieval results. A citation label only identifies a source; it does not prove that source supports the claim.
Can I change embeddings after indexing?
The guide restricts changing the embedding model after parsing in a dataset. Different embedding spaces are not interchangeable; plan migration or re-indexing carefully.
Does it require a paid cloud model?
It supports local model-serving routes as well as external providers. Local operation still needs compatible models and adequate hardware.
What happens if no source answers the question?
Configure an explicit missing-answer response. Leaving the assistant free to improvise can produce claims beyond the dataset.
Can another app use it?
Yes, the project documents an HTTP API. Configure access for the intended application and validate how it handles source references and errors.
The details behind your decision
Explore platforms, inputs and outputs, licensing, and access requirements.
Four useful starting projects
Four useful starting projects
- Technical documentation search: Index a current product manual and ask about prerequisites, limits and procedures. Verify that the retrieved sections belong to the intended model and version.
- Report comparison: Build a collection of related reports and request a comparison with source references. Check that different reporting periods and units have not been treated as equivalent.
- Internal procedure lookup: Offer a controlled assistant over approved process documents. Define how it should respond when an answer is missing instead of allowing improvised policy.
- Document ingestion diagnosis: Use chunk inspection and retrieval tests to find why an existing knowledge application misses answers. Correct the data path before adding more model complexity.
Platforms, inputs, outputs and setup
Platforms, inputs, outputs and setup
The current repository quick-start describes a Go deployment and a supported Linux x86-64 native Docker target. Release instructions can change, so use matching documentation and images instead of mixing a recent development guide with an older production database.
Documents, tabular files, images and slides are supported inputs; parsing quality varies with their structure. The guide warns that an embedding model cannot simply be switched after a dataset has been parsed, because its vectors must remain comparable. Plan a rebuild or a separate dataset when changing that foundation.
Data, permissions and licensing
Data, permissions and licensing
Self-hosting controls where the application and documents are stored, but external embedding or chat providers can receive material required for their work. Inspect the full pipeline, including parsing services and optional tools, before treating a deployment as fully local.
Retain original documents and back up both configuration and stored knowledge. The software license does not grant permission to redistribute a confidential document collection, and an API integration should not accidentally make a private dataset public.
Sources, documentation and help
Official sources and review date
Official sources and review date
Reviewed October 3, 2026. Product facts come from the official sources below. Suggested projects, prompts and review methods are FindGoodAI editorial guidance, not measured performance results.
Your experience can help someone choose
COMMUNITY
Sign in to rate and comment
Tried it? Share your experience
Loading experiences…
Manage my comments
Community and moderation
Saved tools are private. Approved comments are public; edits return to moderation. Comment counts include approved comments and replies. Each person has one rating. New ratings are automatically approved; withdrawn, unapproved or invalid ratings do not count. Share real experiences and avoid spam, private information or personal attacks.
For moderation appeals or data requests: support@findgoodai.com
NEXT STEP
Explore RAGFlow ↗Turn a difficult document into searchable knowledge
Bring a real task and see how it fits the way you work.
Jump to a section
User experiences