Learn and test RAG: a free RAG playground
See exactly which passage retrieval finds for a question, and why it misses the right one. Free, no sign-up, and it runs in your browser. Nothing is uploaded.
Where do you want to start?
Learn how RAG works
8 short steps, about an hour in all. Each one is a page to read or a tool to try, and hands you to the next.
- 1What is RAG5 min
- 2Watch it run10 min
- 3Chunking10 min
- 4Embeddings6 min
- 5Keyword vs meaning6 min
- 6Re-ranking6 min
- 7Testing8 min
- 8Graph RAG6 min
Work on your own RAG
Pick the job. Each one starts you on the right tool, and your document carries across.
Or try it right here
Document
Settings
For a model you run yourself (Ollama, LM Studio and similar) or one you have an API key for, such as Gemini or Claude. Other models must accept OpenAI-style requests and allow requests from . Requests go straight from your browser to this address.
Hint
Open Settings below the document and look at Chunk overlap or Chunking strategy. Watch the Retrieved chunks box as you change them.
Show a fix
Any one of these works: Chunk overlap 10 words (the repeated words keep the sentence whole in one chunk), Chunking strategy Sentence (cuts only at sentence ends), or Chunk size 50 words (moves the boundary). Read why in How to choose a chunk size.
Answer
Nothing in the document matches this question, so there is no prompt to build. Check the document on the left is the one you meant, or try other words.
For a written answer, copy the prompt into any AI chat or bring your own model.
Show the prompt
Retrieved chunks
Wrong or missing chunk? Find out why
Copy a few words from the part of your document that should have been retrieved. You will see why it was missed and which setting fixes it.
Trace: question to answer
| Step | Time | Size | Note |
|---|
Times are measured in this browser tab: the document is read and searched here, and the model is yours. They show where the time goes, not what a production system would take.
Step by step
New: testing retrieval now has its own page, with every setting in one place. Your document carries across.
Evaluation set (test questions)
For each question, enter words that the right chunk must contain. Separate answers that sit in different chunks with ; and phrases that must be in the same chunk with +, for example 30 days + full refund; 14 days. Leave it empty for a question the document cannot answer: it then passes if nothing scores above the minimum score. Tags (optional, separated by commas) give a score per kind of question.
Results
See everything TryRAG can do →
What it is for
- Learn how RAG works by watching every step run on a real document: splitting, vectors, similarity scores, retrieval and the final prompt.
- Check what a RAG system would retrieve from your own PDF or text for a given question, before building anything.
- Tune retrieval: try chunk sizes, overlap, splitting methods and matching options, and measure which finds the right chunk most often.
- Bring your own model (BYOM): connect Ollama, Gemini or Claude and get written answers from the retrieved chunks.
- Teach it: the "Show me" walkthrough and the worked similarity sum are made for explaining RAG to a class or a team.
Questions
Is this RAG playground free?
Yes. TryRAG is free to use online, with no account, no sign-up and no limit on the number of questions.
How large a document can I use?
Up to eight documents at once, each up to 600,000 characters, which is about 100,000 words or 250 pages of plain text. Anything longer is cut at that point, and the page tells you when that happens. The engine is tested with 3,000 chunks. Word matching stays instant at that size; meaning search has to turn every chunk into numbers in your browser first, which takes longer on large documents and slower computers.
Can I ask questions about my own PDF?
Yes. Upload a PDF that contains real text and ask a question. The tool shows the chunks it would retrieve, with page numbers. Scanned PDFs cannot be read.
Does it write the answer?
Not by default. It shows the line of your document that is closest to the question, and builds the prompt, which you can copy into any AI chat. If you run a model on your own computer, or have an API key, you can connect it and get the answer on the page.
Why was the chunk I expected not retrieved?
Under the answer, press "Find out why", enter a few words from the chunk you expected and press Explain. The tool names the reason: the text is cut at a chunk boundary, the chunk ranks below the cut-off, the question uses different words, a minimum score or variety setting removed it, or the words are not in the document text. Where a setting change fixes it, the tool offers that change and applies it for you. In the Evaluate tab, press "Why?" on any question that was not found.
Is my text uploaded?
No. Everything runs in your browser tab. Your documents and questions never leave your device.