Activity-frames – give your AI agent eyes on your day (github.com)

Nocta has introduced a new desktop app feature called "activity-frames," designed to enhance AI agents' understanding of user activity throughout the day. By capturing screen activity locally, the system compiles these interactions into structured episodes, granting agents context for more informed conversations. The feature boasts full privacy since all data processing occurs on-device, eliminating reliance on cloud services or language models, and can summarize an entire day in under a second. The significance of activity-frames lies in its introduction of episodic memory to AI agents, which traditionally relied solely on conversation memory. This capability allows agents to operate with a clearer understanding of a user's recent activity—considering factors like web pages visited, applications used, and duration of tasks—thus supporting more relevant and informed interactions. The implementation ensures reliability through a two-tier data contract, where Tier 1 provides deterministic outputs from raw activity, while Tier 2 allows for inferred contextual insights, reinforcing accountability and transparency in AI interactions. As a tool that records, compiles, and offers contextual awareness—all locally—activity-frames has the potential to redefine user-agent collaborations in workflows.