KnowledgeForgeKnowledgeForge
Your documents, read and understood on your own computer.
KnowledgeForge keeps a library of your PDFs, Office files, web pages, e-books, e-mail and scanned pages, reads every one of them, files each under a category with a summary, and answers your questions from them in plain words, every claim cited, every citation one click from the page it came from. It runs on your machine: nothing is uploaded, no account is needed, and it works without a network once its models are in.
Who it is for
- Engineers and researchers with shelves of datasheets, standards, papers and manuals, who need the answer and the page it is on.
- Professionals whose knowledge is in documents that no search box reads: contracts, reports, specs, scans.
- Anyone who wants an AI over their own files without sending those files anywhere.
What every copy does: KnowledgeForge Free
- Reads what you have. PDFs (scanned ones too), Word, PowerPoint and Excel, web pages, Markdown and text, CSV, e-books, e-mail, pictures, and whole folders and zip archives of them. Files stay where they are, or are copied or moved in, as you choose.
- Files it for you. A local AI writes each document's summary, category, subject and topics; the catalog's folders are those categories, and the queue shows every page being read.
- Finds the passage. Search titles, or the text itself: the passages that answer, ranked and opened at their page.
- Answers with citations. Ask the whole library or one document; follow up; the answer says when the documents do not say. Copy an answer as Markdown with its sources.
- Reads beside you. Open a document at the cited page, or in its own program, and download a copy under its own name.
- Fits the machine. The local AI is picked for the hardware, from a laptop's processor to a large graphics card; search runs on the GPU when there is one, and on the processor when it is full.
- Stays yours. Everything is in a folder you choose; several libraries; signed updates you install when you decide.
Free is free for as long as you like, for a library of up to 50 documents, with one corpus of up to 10 and one research run to try. Every install starts with fourteen days of Pro, with no limits.
What Pro adds
| Pro | What it gives you |
|---|---|
| Research Assistant | Free has one run; Pro researches as often as you like. A topic researched on the web and among papers (OpenAlex, arXiv, Europe PMC, Semantic Scholar), written up as a cited report and then checked claim by claim against its own sources: arithmetic recomputed, unsupported claims marked, speculation flagged, gaps searched again or stated. Research your library alone, the web, or both. Keep the pages worth keeping, with where they came from. |
| Your library in your editor | An MCP server that puts the library inside Claude Code, Cursor, Windsurf and Claude Desktop: search it, ask it, read from it, and have it research for you, without leaving the editor. |
| Watched folders | A scanner's output, a downloads folder, a shared drive: new, changed, moved and deleted files followed by themselves, on network drives too. |
| Libraries and corpora of any size | Free keeps 50 documents and one corpus of 10; Pro has no limit. Named sets of documents (a project's datasheets, a course's papers) that chat, search and research are pointed at. Ask one corpus, or one category, from Chat. |
| Documents you edit | Edit a document in Word, LibreOffice or your editor; KnowledgeForge notices, marks it changed, and re-indexes it in one click, or by itself when you save. |
| Your own models | Chat, summaries or embeddings through Ollama, LM Studio, OpenAI, Anthropic or any OpenAI-compatible server, beside the built-in model; the app always says what leaves the machine. |
| Faster ingestion | Up to eight parse processes on a long PDF, OCR on the GPU, values of your own, and a benchmark that times them on one of your documents. |
| Recordings | Lectures, meetings, interviews and podcasts (MP3, WAV, M4A, FLAC, Ogg and more) transcribed on this computer, nothing sent anywhere. Ask them like any document: a citation says "at 12:34" and plays the recording from that moment. |
| More formats | CHM help files and the older .doc, .ppt, .xls and .rtf, converted with 7-Zip or LibreOffice. |
Pro is one payment for one license key for two computers, bought on the KnowledgeForge website and activated in the app; the app takes no payment itself. For a third computer, another Pro key. When a trial or a license ends, the copy is Free and keeps everything it made: a library over 50 stays searchable and takes new documents once it has fewer; of several corpora the first in the list is the one used, and the others can be deleted; reports stay readable and exportable.
Home-Lab Server and Enterprise
Home-Lab Server is on sale: KnowledgeForge on an always-on machine of your own, reached in a browser, with one key for the server and five client desktops, each with everything Pro has; more desktops added to the same key, each one payment. Until the desktop bridge ships — the app using the server's GPU and library — it sells at an early-adopter price, for the server and for each desktop added, and the bridge comes with it. Enterprise (a department, with its own support) is licensed by contract. Home-Lab Server and Enterprise are the licenses that connect the desktop app to a KnowledgeForge server; Free and Pro stand alone.
Every edition is one payment, not a subscription. Every license covers every 1.x release; 2.0 will be a paid upgrade at a discount for anyone holding a 1.x key. Every sale is final: the 14-day trial is the evaluation. Faults in the product are fixed free for every edition; help beyond that is a flat-priced support ticket.
Privacy, plainly
Documents, index, summaries and conversations stay on the computer, in folders you can back up or copy. The app uses the network for four things: downloading its models once, the license server (the trial recorded once at the first start, the license buttons you press, and a daily check that a held license still stands), the update check, and, only when you ask, research searches and any model you point a job at. The privacy policy names every host.
Runs on
Windows 10 and 11 (x64), by installer or from the Microsoft Store, and Linux (x86_64 AppImage or .deb); macOS coming in November. About 10 GB of disk for the components and the search models, plus 1.4 to 19 GB for the local AI depending on the machine. An NVIDIA card is used when there is one; none is required.