February 24, 2026

Moments and Curiosity Threads

Introduced Moments, a new core feature for bookmarking conversations and seeding long-term curiosity threads, and refined the associated front-end UI.

Moments & Curiosity Threads

The centerpiece of today’s work was the initial implementation of “Moments.” This is a major new feature that allows pinning important messages from a conversation, effectively creating a bookmark that Chalie can semantically recall later. This is more than just storage; it’s the foundation for what we’re calling “Curiosity Threads.”

These threads are long-term, self-defined goals or topics of interest that can emerge from user interactions. The system can now autonomously seed these threads from saved Moments, enabling it to pursue lines of inquiry or bring up relevant context in future conversations. This was a massive architectural lift, touching dozens of files across the cognitive stack — from new database migrations and LLM prompts to core services for context relevance, cognitive triage, and innate skills like ‘focus’ and ‘goal’. We also added the necessary LLM agent configuration for the background “moment-enrichment” process, which will handle summarizing and titling these saved snippets.

Moment Card UI/UX

With a new core feature comes a lot of front-end work to make it usable. We spent a good portion of the day polishing the “Moment card” UI. Pinned messages now correctly render markdown, making them much more readable. We also tackled a text repetition issue by making parts of the message collapsible and improving the title fallback logic.

Most importantly from a UX perspective, we added a proper “forget” flow. Instead of a moment being instantly deleted, tapping ‘Forget’ now puts the card into a pending state for 10 seconds, giving the user a crucial window to undo the action. This makes the experience much safer and less stressful.

Backend Refinements

Finally, a couple of backend improvements went in. We replaced the old, ambiguous scheduler types reminder/task with the more descriptive notification/prompt, which better clarifies their behavior and presentation. We also pushed a small but important bug fix for multi-step tool and skill calling to ensure more reliable complex actions.