v1.7.4

jot v1.7.4

This release is a focused production backport of assistant memory to the 1.7.3 line. The goal is to make jot assistant better at remembering useful personal context, recalling it at the right time, and maintaining memory over time without pulling in the broader unreleased assistant work from main.

What's new

• Added first-class assistant memory tools:
  • memory.remember
  • memory.search
  • memory.recall
  • memory.inspect
  • memory.update
• Upgraded assistant memory from simple scaffolding to a richer local memory model with:
  • observations
  • inferred facts
  • confidence
  • verification
  • memory kinds
  • lifecycle buckets
• Added memory recall injection in the assistant runtime so relevant stored context can be brought into later turns automatically.
• Added post-turn memory consolidation so the assistant can store useful facts and situations from conversations over time.
• Added memory lifecycle behavior for:
  • reinforcement of repeated facts
  • expiry of replaced active states
  • scheduled-to-active-to-expired rollover
  • confidence decay for stale weaker memories
• Improved semantic normalization and dedupe so similar facts are more likely to reinforce instead of duplicating.

Notes

• This release is intentionally scoped to memory. It does not include the newer daemon, WhatsApp/runtime expansion, proactive feed, or broader chat UX work from the current development branch.
• Memory remains local-first and stored on disk in the assistant config area.
• Assistant memory quality is materially stronger than 1.7.3, but some inferred facts can still vary based on model output quality.

Release notes

• GitHub release assets are built locally for v1.7.4.
• This release was cut from the backport/memory-1.7.3 branch instead of main, because main contains unrelated unreleased development work.