Links are not yet activated.
To activate, add a link back to submitpr.org from your website and contact @jaycosta on Telegram,
or pay via Solana (from $19.95) for instant activation.
ReadyCode AI has released ReadyCode Reader, a free, open-source Jev AI-powered tool that helps AI assistants answer specific questions from large PDF, Word and Excel files without repeatedly reading the entire document.Reader is available as a Model Context Protocol (MCP) server for Claude Code, Cursor, Codex and other MCP clients, and as a browser demo. It reads the source file on the user's computer, performs local keyword search, and returns only the passages relevant to each question together with page, section, paragraph, sheet or row citations.
The project uses TypeSafe's Jev decision model to check whether retrieved evidence is relevant and sufficient, whether passages conflict, and whether text contains instruction-like content that should be treated as data rather than followed. For spreadsheet questions, Reader can perform exact counts, distinct-value checks, rankings, totals, averages, minimums, maximums and lists over every row instead of asking a language model to estimate the answer.
In ReadyCode AI's reproducible comparison using a 200-page NASA PDF and eight questions, both standard Codex and Codex with Reader answered 8 out of 8 questions correctly. The standard workflow used 913,468 tokens and took about three minutes. The Reader workflow used 41,997 tokens and took about 20 seconds. Benchmark question files, checking scripts and full results are published in the project's GitHub repository. These figures are ReadyCode AI's own runs on the tested files, not an independent study.
ReadyCode Reader also includes code-checked tests using a 263,000-token Word report and a 100 MB Excel workbook containing about 1.1 million rows. The project is designed for precise questions rather than whole-document summaries, and image-only pages require a separate vision tool.
Privacy is explicit: the full file stays on the user's computer and is never uploaded to ReadyCode. For each question, the question and a shortlist of candidate passages are sent to an OpenRouter-hosted Jev decision model using the user's own API key. Exact spreadsheet calculations run locally.
Reader is licensed under Apache-2.0. Developers can install it from npm with npx -y @readycode/reader, inspect the source and benchmarks on GitHub, or try the browser version through the ReadyCode Reader website.
About ReadyCode AI
ReadyCode AI develops free tools that help people use AI more efficiently while keeping evidence visible and verifiable. ReadyCode Reader is available at https://readycode.ai/reader and https://github.com/ReadycodeAI/readycode-reader.