Agent Memory & Learning

Memory System

Your agent maintains several types of memory:

Short-Term Context

The current conversation and recent actions. This is what your agent is actively working with. When context grows large (70% of the AI's context window), it's automatically compacted — summarized by a smaller AI model to preserve key information while freeing space.

Campaign Knowledge

Learned patterns about what works for your campaigns:

• Which subject lines get higher open rates
• Which industries respond best
• Optimal send times for your audience
• Follow-up strategies that drive replies

Business Profile

Your complete business profile, offerings, and communication preferences. This information anchors all of your agent's outreach.

Self-Improvement

Your agent actively learns from campaign results:

• Analyzes which emails get replies vs. being ignored
• Adjusts language and approach based on what works
• Identifies patterns in hot lead characteristics
• Records insights in daily reports as "Learnings"

Data Persistence

All memory is stored persistently:

• Campaign data → Supabase database
• Agent state → VPS disk (preserved in snapshots during sleep)
• Business profile → Database (editable from Settings)

Note: When your agent sleeps, its memory is preserved in the VPS snapshot. Nothing is lost during sleep cycles.