Active memory

Active Memory is the deep-recall lane for eligible conversational sessions. The default escalate mode runs its blocking recall sub-agent only when the message asks about the past and the deterministic memory lane found no strong trusted trigger match. This keeps ordinary replies fast while preserving a deeper search path for prior decisions, conversations, and temporal or multi-hop questions.

Flat retrieval is strongest for direct fact matches and weaker on temporal and multi-session questions. LongMemEval (arXiv:2410.10813) measures that gap, while the PrefEval benchmark highlights the value of preference-adjacent reminders. Escalation by default spends the blocking model call where those harder recall shapes are actually present.

This page is an index. Active memory is documented on nine pages, one per reader job. Open the page that matches your task.

Page	Read it when
Enabling active memory	You are turning active memory on, through the per-agent product setting or the advanced plugin config.
How active memory works	You want the escalation flow, or you are checking whether a conversation is eligible.
Session controls	You want to pause it for one conversation, or see what it injected.
Query modes, prompts, and models	You are trading recall quality against latency, or picking a recall model.
Memory tools	You are choosing recall tools, or wiring active memory to built-in memory, LanceDB, or Lossless Claw.
Advanced overrides and transcripts	You need a prompt or thinking override, or blocking sub-agent transcripts on disk.
Configuration reference	You need the type, range, or default for an active memory key.
Recommended setup	You want a starting configuration, or the cold-start grace budget after a v2026.4.x upgrade.
Troubleshooting active memory	Active memory is not running where you expect, or recall is slow, empty, or policy-disabled.

Where each section moved

Every section heading from the previous single-page version keeps its anchor here, so an existing link such as /concepts/active-memory#lossless-claw still resolves. Each entry points at the page that now holds the content.

• Remember across conversations
• Advanced Active Memory quick start
• How it works
• When it runs
  • Session types
• Session toggle
• How to see it
• Query modes
  • Query mode: message
  • Query mode: recent
  • Query mode: full
• Prompt styles
• Model fallback policy
  • Speed recommendations
    • Cerebras setup
• Memory tools
  • Built-in memory
  • LanceDB memory
  • Lossless Claw
• Advanced escape hatches
• Transcript persistence
• Configuration
• Recommended setup
  • Cold-start grace
• Debugging
• Common issues
  • Registered recall tools return status=policy-disabled
  • Embedding provider switched or stopped working
  • Recall feels slow, empty, or inconsistent
  • First recall after gateway restart returns status=timeout

Related pages

• Memory architecture
• Memory Search
• Memory configuration reference
• Plugin SDK setup