Ambient Context adds text-based activity logging to macOS for AI agents
The open-source tool creates a local, searchable record of user activity to provide long-term memory for LLM agents.
Developer dragthelake has released Ambient Context, a macOS menu bar application that creates a continuous text-based record of a user's digital activity. The tool provides a durable memory source for LLM agents, such as Claude Code, allowing them to reference historical work without relying on invasive recording methods.
Unlike traditional activity trackers that utilize screen recording or Optical Character Recognition (OCR), Ambient Context leverages the macOS Accessibility API to read text directly from the focused window every few seconds. This captured data is stored locally as plain Markdown files, with one file generated per day in a folder selected by the user. To protect sensitive information, the application includes built-in redaction for private browsing and password managers, as well as pattern-scrubbing for credentials and API keys. The app requires macOS 14 or newer on Apple Silicon and is currently unsigned, meaning users must build it manually using Node, Rust, and Xcode Command Line Tools.
The shift toward agentic workflows
The release comes as the industry shifts toward "agentic workflows," where AI agents are expected to operate with a deeper understanding of project progress and historical context. Currently, providing this context often requires manual input or high-overhead tools like constant screen recording, which present significant privacy risks and storage challenges. By converting active window content into a structured, text-only log, Ambient Context allows an AI agent to answer specific questions about past activity—such as what was worked on a specific day—through simple local file access.
A privacy-first approach to memory
This approach proposes a privacy-centric alternative to "screen-memory" tools. By avoiding cloud-based processing and OCR, the tool ensures that sensitive work data remains on-device. This solves a critical bottleneck in the AI experience: the limited context window of LLMs. Instead of attempting to feed massive amounts of raw data into a prompt, the tool provides a searchable, deduplicated history of actual digital activity that can be queried as needed.
Technical limitations and outlook
While the tool offers a streamlined way to log activity, it faces technical hurdles based on how different macOS applications render text. The developer notes that while Terminal.app and iTerm2 function correctly, GPU-rendered terminals like Kitty and Alacritty expose little to no text to the Accessibility API. Additionally, users may notice slightly glitchy window-resize animations in Google Chrome while the app is active. Future adoption will likely depend on whether the developer provides signed binaries to simplify installation for non-technical users.