plai.chat
plai.chat
Add a link or a file to any plai.chat conversation — Markdown, text, JSON, YAML or CSV — and ask questions. Answers cite your sources — a lightweight RAG (retrieval-augmented generation) built into every chat. Not stored on our servers.
Open PLAI.chat →Free account required · up to 3 sources per chat · 2 MB each · 300+ models
RAG (retrieval-augmented generation) means the AI first looks up the relevant parts of your own documents and then answers from them, citing where each fact came from. Usually that needs an embeddings pipeline and a vector database. In plai.chat it is a simple RAG with nothing to set up: add a source to the chat and ask. Running it for a team or a small business? See simple RAG chat for everyone and small businesses.
A short document is given to the model whole, so nothing is missed. Under the answer it is listed as read in full.
A bigger file is split into passages and indexed. For each question, the most relevant passages are retrieved and sent to the model, which cites them with numbered references. When the source contains URL: lines — as an llms-full.txt does — the Sources list links straight to the right article or section.
Tested with Polish, including questions typed without diacritics. Asking in the document's language gives the best matches.
Combine knowledge sources with per-chat custom instructions — for example a tone, an answer length or a role.
| Format | Extensions | Typical source |
|---|---|---|
| Plain text / Markdown | .txt, .md | llms.txt, llms-full.txt, README, notes, policies |
| JSON / JSONL | .json, .jsonl | API responses, exported data, configs |
| YAML | .yaml, .yml | OpenAPI specs, configs |
| CSV / TSV | .csv, .tsv | Price lists, catalogues, spreadsheet exports |
| Limits | Up to 3 sources per conversation · up to 2 MB per source · link or upload | |
| Not yet | HTML web pages and PDF files | |
A link must point to a text file, not a web page: for example a site's /llms.txt or /llms-full.txt, a raw Markdown file, or a CSV export.
Link a product's llms-full.txt and ask how a feature works. Answers cite the right section and link to the original article.
Upload a CSV export of prices or products and ask “which items under $50 ship in 24 h?” — every retrieved chunk keeps the header row.
Answer questions from a returns policy, terms or an internal FAQ written in Markdown, grounded in that document.
Attach an OpenAPI spec or a JSON/YAML config and ask which endpoint or field does what.
Quiz yourself on your own lecture notes or a reading summary, and let the model explain the parts you got wrong.
Drop in a text or Markdown export of meeting notes and ask for decisions, owners and open questions.
Ask about onboarding steps, leave rules or tooling from a handbook saved as Markdown or plain text.
Add two documents — two price lists, two versions of a policy, two vendors' docs — and ask what differs.
Support agents can answer customer questions from a knowledge file, combined with per-chat instructions on tone and length.
Ask in the language of your documents — retrieval is multilingual and tested with Polish, even for queries typed without diacritics.
plai.chat does not store your documents or your chats permanently on its servers. Uploaded files are kept in your browser (best effort — not in private windows or when site storage is blocked); links are stored as links. The server holds the text and index in memory only and rebuilds them after a restart — links are re-fetched, uploads re-sent from your browser; if the browser copy is gone, you are asked to upload the file again.
Zero Data Retention mode applies to the embedding requests too. The text is processed by the embedding provider (via OpenRouter), and to answer, the relevant text is sent to the model you choose.
Embedding and indexing costs are covered by plai.chat. Answers use normal pay-per-use pricing: on paid models the source text sent with each message counts as input tokens, so messages with sources cost more. Free models stay free.
Yes. Sign in with a free account, open any chat and add up to 3 knowledge sources from the composer's “Chat context” (book) button. It works with free models too. Indexing and embedding are free to you — covered by plai.chat. Answers use normal model pricing: on paid models the source text or passages sent with each message count as input tokens, so a message with sources costs more than one without.
.txt, .md, .json, .jsonl, .yaml / .yml, .csv and .tsv — uploaded as a file or fetched from a link — up to 2 MB per source. HTML web pages and PDF files are not supported yet, so a link must point to a text, Markdown, JSON, YAML or CSV file such as llms.txt, llms-full.txt, raw Markdown or a CSV export.
Small sources are read in full by the model and listed as “read in full” under the answer. Larger sources are indexed, and for each question the most relevant passages are retrieved and cited with numbered references [n], plus a Sources list that links to the right article or section when the source includes URL lines (as llms-full.txt files do).
No. Documents are not stored permanently on our servers. Uploaded files are kept in your browser (best effort; not in private windows or when site storage is blocked) and links are stored as links. The server holds the text and index in memory only and rebuilds them automatically after a restart — links are re-fetched, uploads re-sent from your browser; if the browser copy is gone you are asked to upload the file again. To answer, the text is processed by the embedding provider (via OpenRouter) and sent to the chat model you choose. With Zero Data Retention mode on, the embedding requests use ZDR too.
Any chat model in plai.chat — 300+ models, free tier included. You can switch models and ask follow-up questions in the same conversation. Knowledge sources are not used for image generation.
Yes. Retrieval is multilingual and was tested with Polish documents and questions, including questions typed without diacritics. Asking in the same language as the document gives the best matches; questions in another language may find fewer relevant passages.
Yes. Each chat can have its own custom instructions (for example “answer as a support agent, briefly, in German”) alongside up to 3 knowledge sources.
Up to 3 sources per conversation and 2 MB per source; a very large file can still exceed the 4,000-passage limit. Each account can index up to 10 new large sources per day, and adding sources is capped per hour; if the service-wide daily indexing budget runs out, retrieval is temporarily unavailable. No PDF or HTML pages yet. A free account is required. If free models are busy, a backup model may answer without your sources.
Free account. 300+ models. Answers with sources. Not stored on our servers.
Open PLAI.chat →