What is HaroCue?
HaroCue is a local-first personal desktop AI assistant. It turns permitted work context into personal memory with sources, offers Cue suggestions and / questions where you write, and supports daily, weekly and monthly reviews.
Where does memory come from?
With recording enabled and system permission granted, the app captures accessible text from allowed foreground windows. Optional, authorized local OCR can supplement it. Your configured model extracts clues, preserves sources and times, and connects related topics. You can also add a memory manually.
Are extracted memories confirmed facts?
No. Automatically extracted observations are assistant inferences. You can inspect the original, correct or forget them. Ordinary Cue uses only explicitly confirmed memory facts. Inferences used in answers still need checking.
Does data stay local?
Memory is primarily stored on your device. Cloud models and online tools may receive relevant text and memory needed for a request. Optional OCR processes screenshot originals locally; recognized text may enter model requests. Local Ollama is also supported.
Which systems are supported?
The desktop trial checked on September 9, 2026 is version 0.2.12. Recent desktop features are mainly in local trials on Apple Silicon macOS. Public packages may predate the website demos. Check Downloads and release notes for Windows, macOS and Linux build availability and capabilities.
How does pricing relate to model costs?
The local core is free and can use your own configured model. Plus (proposed $9/month), Pro (proposed $19/month) and annual plans remain draft hosted service proposals; the website does not currently collect payments. Model providers may bill separately.
How does it work with other AI tools?
Codex, Cursor or Claude Code can retrieve relevant local context when you explicitly invoke HaroCue MCP. Installing it does not automatically share all memory; retrieval scope and detail permissions remain controlled.