Kortyx is a proactive AI assistant that learns how you work, captures context from desktop and browser activity, and can suggest and carry out tasks with your approval. It is designed for people who want an AI agent that can use their work history, drafts, and pages to help finish routine or incomplete work.
Kortyx learns from the work you already do, then uses that context to suggest tasks it can take off your plate and complete for you. It is designed around capture, memory, delegation, and execution rather than one-off chat prompts.
The product is available as a desktop app for macOS and Windows, with optional Chrome and Telegram connections. According to the site, it can observe desktop and browser activity with permission, answer from memory, run browser sessions, execute code in an isolated sandbox, and wait for approval before sending, submitting, or spending anything consequential.
Kortyx records work context with permission and turns it into structured memory it can use later. The site says it can capture apps, pages, meetings, drafts, and browser activity while redacting sensitive information such as passwords, payments, and private browsing on the device.
Instead of waiting for a prompt, Kortyx can notice unfinished or stalled work and send a proposal through desktop chat or Telegram. It only pings when it has a specific task it believes it can help finish, and it shows what the suggestion is based on.
When a task is accepted, Kortyx can plan the work, use a browser session, fill forms, and interact with pages. More sensitive steps are approval-gated, so the user remains the final check before anything is sent or submitted.
The product says it can write code for scripts, spreadsheets, and file processing in a secure sandbox. This keeps the work separate from the user’s machine while still allowing it to handle more technical tasks when needed.
Paid plans include deeper memory access and full-depth grounded chat over private context. The site also describes Recall history, which surfaces past context that the assistant has learned from previous work.
Kortyx mentions multi-agent MCP access to the same memory and the ability to save repeated tasks as skills. It also says repeated work can become scheduled skills that the assistant can rerun later.
A user can rely on Kortyx to find drafts, notes, and recent context, then propose a concrete next step such as finishing a comparison, updating a decision page, or sending a follow-up that was left unsent.
Kortyx can open pages, compare current information from browser sessions, and prepare actions that require navigation or form entry. The site shows it being used for vendor pricing comparison and publishing a recommendation after review.
Telegram support lets the assistant message the user about useful tasks it can complete and receive approval remotely. This is meant for times when the user is away from their desktop but still wants to review or authorize work.
When a task happens often, Kortyx says it can convert that work into a skill and rerun it on a schedule. That makes it more relevant for recurring routines such as updates, checks, or periodic follow-ups.
For jobs that go beyond chat and browser clicks, Kortyx says it can run code in a secure sandbox to process files or automate spreadsheet-style tasks while keeping execution off the local computer.
The site says Kortyx has a desktop app for macOS and Windows. It also offers a Chrome extension for browser context and Telegram linking for delegation and proactive pings.
Kortyx can plan and execute work, but external actions and higher-risk steps are approval-gated. The site specifically says sending, submitting, or spending anything waits for user approval.
Yes. The pricing FAQ says capture can be paused at any time, and sensitive contexts such as passwords, payments, and private browsing are redacted on the machine.
The pricing page shows a Free plan and a Pro plan at $20 per month. It also says Kortyx is not priced by AI tokens; you bring or connect your own AI provider in the app.
The site says Pro unlocks the deeper parts of the product, including full Recall history, ongoing proactive delegation, full-depth grounded chat over private context, and multi-agent MCP access to the same memory.
Traffic-Daten dienen nur als Referenz.
| Mai | 183 |
|---|---|
| Jun | 0 |
| Jul | 2912 |
Traffic-Analysen sind noch nicht verfügbar.
Traffic-Analysen sind noch nicht verfügbar.
