Plugs into the AI tools you already use
Claude
Cursor
Cline / Roo
OpenCode
Codex
Kimi CLI
Hermes
Pi
OpenClaw
Any MCP client
Wikori is a desktop app that turns the documents, emails and links piling up around you into one organised knowledge base — kept on your own machine, in plain readable files, and instantly searchable by you and by Claude, Cursor, or whichever AI assistant you use.
macOS · Windows · Linux · One-time $99 — no subscription
Plugs into the AI tools you already use
Claude
Cursor
Cline / Roo
OpenCode
Codex
Kimi CLI
Hermes
Pi
OpenClaw
Any MCP client
Three steps, then it runs on its own. No tagging, no filing, no folder discipline — the app does the boring part.
Drag a contract, a report, a photo of a whiteboard. Paste a link or a YouTube talk. Forward an email. Point the crawler at a site and it collects every page. If it carries knowledge, Wikori takes it.
Text is pulled out of every format, then your AI model writes a title, a summary, the people and companies involved, the topics and a confidence score — and saves it all on top of the file. Use a cheap hosted model, or run it entirely on your own machine with Ollama. One short call per document, once.
"Brief me on this client before the call." Your assistant searches the vault itself, reads only what matters, and answers with your material — not a guess. Nothing to paste, nothing to re-explain, and nothing more to pay: the index and the search live on your machine.
Not a cloud account you rent. Not a database you can't open. A vault is an ordinary folder full of ordinary Markdown files — sitting on your machine, backed up by whatever you already use, readable in twenty years by anything that can open a text file.
--- title: Acme — Master Services Agreement 2026 summary: Three-year MSA. 90-day termination clause, annual price review capped at 6%. entities: - Acme Corp (company) - Maria Alves (person) tags: [contract, vendor, pricing] source_type: contract source_authority: high confidence: 0.94 memory_tier: semantic --- # Original document text follows…
Press ⌥ ⌥ — both Option keys — from anywhere. A floating note appears over whatever you're doing. Write. Assign a vault. Vanish. Your AI processes the rest.
Several clients, several projects, a river of incoming material. Wikori keeps it separated, structured and ready — without you filing a thing.
One per client, project or department. Each fully isolated — its own files, its own pipeline, its own email tag. Switch in a single click and your AI follows.
An interactive graph of every person, company and topic in your vault, and the documents they connect. Click a node and follow the thread — you'll find links a folder never shows you.
Paste a URL to capture the page. Paste a YouTube link to capture the transcript. Or point the crawler at a site, review what it found, and ingest the lot in one batch.
Forward newsletters, client updates or internal memos to a dedicated mailbox. Wikori collects them from trusted senders only and files them into the right vault, automatically.
Whiteboard photos, scanned receipts, diagrams, slide decks, spreadsheets. Office files are converted to text; images are read by an AI vision model. All of it becomes searchable.
Press ⌥ ⌥ from anywhere. A floating note appears above every app. Write, pick a vault, send — without breaking focus once.
Exact terms match exactly — a contract number, a tax ID, a part code. Turn on the optional semantic layer and meaning matches too, in 50+ languages, computed on your machine at zero token cost.
Your signed HR policy outranks a blog post that happens to use the same words. Every entry carries a confidence score and an authority level, and both shape what surfaces first.
Lives in your system tray, watching the drop folder. Close the window and it keeps working. Send files to it straight from your file manager or the command line.
Knowledge bases rot quietly: duplicates pile up, the index falls behind, files sit unprocessed. Wikori schedules the maintenance per vault and does it while you're not looking — daily, weekly, or whenever you say.
Scans the vault for anything that never made it through the pipeline and pushes it back in. Nothing sits forgotten in a corner.
Regenerates the vault's index and schema so search, the Knowledge Map and your AI agents are always looking at current reality.
Compares captured pages by source and content fingerprint, keeps the newest copy of each, and deletes the rest. Re-crawling a site never bloats your vault.
Re-visits a saved crawler profile to pick up new pages from a source you track — a competitor's blog, a regulator, a documentation site.
Each vault keeps its own schedule and its own log of what ran, when, and what it did. Run any routine by hand at any time.
Wikori exists because its founder was making decisions across hundreds of contracts, transcripts and filings — and re-briefing his AI from scratch every single session. If that sounds familiar, it was built for you.
Vendor contracts, compliance reports, strategy transcripts, pricing tables, regulatory updates — scattered across a decade of folders and inboxes. Wikori makes them one thing you can ask questions of. Before a meeting: "brief me on this client." Before a decision: "how do these terms compare with what we already signed?" Answers come from your own material, with the most reliable sources first, and the decisions you've already made stay made.
Every client gets their own vault — proposals, deliverables, call transcripts, competitor research. Switch vault and your AI switches context with you, with no chance of one client's material bleeding into another's. When the phone rings unannounced, you are not scrolling through folders looking for what you promised them in March.
Papers, articles, podcasts, conference talks, book highlights — you take in far more than you can ever file. Wikori absorbs the lot and shows you the overlaps: the same company turning up in three unrelated papers, a regulation cited in both a proposal and an audit. The Knowledge Map shows what a folder of PDFs never can.
Internal docs, architecture decisions, client requirements, incident write-ups — readable by Cursor, Claude Code or Cline while you work. Your agent stops asking you to re-explain last quarter's decision and starts citing it, and it writes what it learns back for the next session. One connection, no copy-paste.
There is no Wikori server holding your work, because there is no Wikori server. You never create an account, never sign in, and never trust us with a contract, a client name or a private note. Everything lives on your machine in plain Markdown you can read, edit and back up yourself.
Every enriched entry is a plain .md file with YAML frontmatter. Open it in any editor, grep from the terminal, read it in Obsidian. No proprietary format, no lock-in.
Wikori never rewrites your source documents. PDFs, Word files, images — they stay exactly as you dropped them, sitting untouched beside the enriched entries it generates.
Your vault is a normal folder. Searching it, browsing it and reading it need no connection at all — only enriching a brand-new file calls out to the AI endpoint you chose.
No sign-up, no login, no profile, no telemetry pipeline pointed at us. Buy the app, run the app. We never see your vault, and we couldn't hand it to anyone if we were asked.
Point Wikori at Ollama or LM Studio on your own machine and no document text ever touches the internet — not even during indexing. Prefer a hosted model? Any OpenAI-compatible provider works, and a small cheap model is plenty for this job.
The index and the vector database sit in your own storage. Once a document is indexed, searching it, re-reading it and serving it to your AI agent — a hundred times a day — costs exactly nothing.
Point your vault at a synced folder and it just works — Google Drive, iCloud, OneDrive, Dropbox, pCloud, Syncthing, Nextcloud — whatever you trust. Wikori doesn't care.
Close the tab and a chatbot forgets you existed. Connect it to Wikori and it doesn't: it reads your vault directly, and it writes back what it learns — the decision you made, the mistake it made, the pattern it spotted. Tomorrow's session begins where today's ended.
{ "mcpServers": { "wikori": { "command": "npx", "args": [ "-y", "mcp-remote", "http://127.0.0.1:8787" ], "env": { "AUTH_TOKEN": "your-mcp-secret" } } } }
Raw notes from a single session get consolidated into session summaries, and the summaries that hold up become settled facts. Your agent stops re-litigating questions you answered in March.
Raw observations from one session — "client prefers fixed-price"
A week of those, consolidated into what actually happened
The durable fact, trusted and cited from then on
A knowledge base you rent is a knowledge base you can lose. Wikori is a desktop app you buy — no seats, no tiers, no monthly bill, no feature held hostage behind an upgrade.
Lifetime licence for one person. Every feature, every update, every vault.
Get WikorimacOS, Windows and Linux · activate on your machine, transfer it whenever you change machines
Wikori reads your documents with your AI endpoint, and you choose what that is. Run a model locally through Ollama or LM Studio and indexing costs nothing and never leaves your machine. Prefer something hosted? Any OpenAI-compatible provider works — OpenAI, Anthropic, OpenRouter, Groq, DeepSeek, Mistral and the rest — and this job doesn't need a flagship model, so the small cheap tiers do it well. You pay your provider directly; we never sit in the middle and never mark it up.
Indexing is a one-off cost per document. From then on, Wikori serves that knowledge to your AI agents for nothing — the index, the search and the vector database all live on your own disk. Ask a hundred questions a day, re-read the same contract every morning, connect five different assistants: none of it produces an API charge.
Your vault is a folder of Markdown files. Delete the app and every document, summary and note you built stays exactly where it is, readable in any text editor. There is nothing to export because nothing was ever locked in.
Multi-seat packs are available — tell us how many people you need to cover.
Your files stay in a folder on your disk — Wikori has no server to send them to. The one exception is deliberate and visible: to enrich a document, its text is sent to the AI endpoint you configured, exactly as it would be if you pasted it into that provider's chat. Choose a local model through Ollama or LM Studio and even that stops happening — indexing then runs entirely inside your own machine, with nothing crossing the network. Searching, browsing and reading your vault work fully offline.
You point Wikori at an AI endpoint of your choosing, and you pay that provider directly — or nobody, if you run a model locally through Ollama or LM Studio, which is fully supported and keeps indexing entirely offline. If you'd rather use a hosted model, any OpenAI-compatible provider works, and this task doesn't call for a flagship: the small cheap tiers summarise and tag documents perfectly well. Indexing is one short call per document, so a few hundred files typically lands in low single-digit dollars — once.
No. Once a document is indexed, serving it costs nothing. The index, the ranked search and the vector database all run on your own machine, so searching your vault and handing the results to an AI agent produces no API charge at all — however often you do it. The only tokens you ever spend on Wikori's side are the one-off calls that index new material. Whatever your assistant charges for its own conversation is between you and them, exactly as it is today.
No. $99 once, for a lifetime licence, including updates. There's no renewal that switches the app off and no tier that unlocks the features you actually wanted.
To use the app: no. You create a vault, paste your API key, and drag files into a folder. Connecting an AI assistant means copying one block of JSON out of Wikori and pasting it into your assistant's settings — the app generates it for you, filled in and ready.
Anything that speaks MCP: Claude Desktop and Claude Code, Cursor, Cline and Roo Code, OpenCode, Codex, Kimi CLI and others. For assistants without MCP support, Wikori generates a prompt that points them at your vault's index instead.
Obsidian, yes — point a vault at your Obsidian folder and the enriched Markdown opens natively. Any folder you already sync through iCloud, Dropbox, Google Drive, OneDrive, Syncthing or Nextcloud works the same way. Wikori adds a metadata block to files; it never takes them over.
They're kept, untouched, beside the enriched entry. Wikori writes new Markdown; it doesn't rewrite your PDFs, spreadsheets or images.
API tokens and mailbox passwords are encrypted at rest with AES-256-GCM, using a key derived from your machine — they're never written in plain text and don't travel with a copied settings file.
Everything you've read, signed, discussed and decided — in one place, on your machine, ready the moment you or your AI need it. One-time $99 for a lifetime licence. macOS, Windows and Linux.
No subscription. No account. Bring your own AI endpoint.