Image: GitHub
Why it mattersA desktop search tool that reads inside files can replace a paid app. Kensaku adds a read-only MCP server. One open-source install gives the user and the agent the same index.
You forget the file name. You remember the error message inside the screenshot. Kensaku is a new open-source desktop app that lets you search by the thing you remember. The GitHub repository, yjid/kensaku, reached 221 stars in its first 7 days since 24 October. It runs on Windows, macOS and Linux. The licence is MIT.
A short task test, written in plain words. Finding files by name is a solved task. Reading inside PDFs, DOCX, screenshots and code, locally and free, is not. That is what Kensaku adds. It also exposes the same index to AI agents through MCP.
What the README claims, with numbers
The README states a short set of measured results. Filename search averages 1.3 milliseconds across 100,000 files, with a best of 0.5 and a worst of 2.4. A first-run index of 101,101 files takes 184 seconds and peaks at 183 MB of memory. The project ran that benchmark on a release build on Linux with 8 virtual CPUs and an SSD. The harness is at docs/BENCHMARKS.md in the repository.
The README also compares Kensaku to plocate and fd on the same folder. Kensaku replies in 1.3 milliseconds on average, plocate in 34, and fd in 109. The README is clear that this is not a perfect match. Kensaku runs inside a long-lived app, and plocate and fd start a new process for each search. Spotlight, Everything and Listary have not been measured yet, the author writes.
What agents can do with Kensaku
A read-only MCP server comes with the CLI. The repository shows the config to add kensaku-cli as an MCP server in Claude Code or Cursor. The agent can then search the same local index the user searches. It can read file contents, and the result size is bounded. The agent cannot change files. Reindexing is off unless the user turns it on. For HTTP clients, kensaku-cli serve listens on loopback only. It requires a bearer token from the user's config directory.
The README says it respects .gitignore and .cursorignore by default. Credential files such as .pem, .key and .pfx are excluded from the index. The project calls this a local-first promise, and notes that no bytes leave the machine.
Where it fits in a working day
A reader who already uses Spotlight on macOS, Everything on Windows or Alfred as a paid tool knows the task. Kensaku adds three things those apps do not do together. Text inside documents and PDFs. Text inside screenshots and image files, with OCR. Natural-language time queries such as "last week" or "yesterday". The query language also supports precise filters: type:pdf, ext:md, in:"~/Notes", modified:7d.
The install size on Windows is 9 MB. Idle memory in the tray is 20 MB of private memory. The app trims memory the moment the user hides it. Builds are not code-signed yet, so macOS and Windows show a first-run warning. The README walks through SmartScreen and macOS Privacy settings. The project is on version 1.0.6. Six things are in the current release: name search, text search, meaning search, typo-tolerant matches, incremental indexing, and agent access over MCP. Image OCR uses PP-OCRv6 by default. Intel Macs fall back to Tesseract.
Source
yjid/kensaku, GitHub.
This item was written by an AI system from the linked source. Reveneau is responsible for what it publishes.
Get AI News in your inbox
New developer tools, model and agent releases, and how teams are actually using them to release software. Short, and only when there is something worth reading.