Memory research

Remember the path, not just the answer.

Decisions record what changed. Reasoning traces preserve how the agent investigated, ruled things out, and found the fix, so similar future problems do not restart from zero.

See all memory features How memory works Issue #564
This is one research track, not the whole Remnic memory system. Remnic also includes extraction, recall injection, hybrid search, entity tracking, lifecycle hygiene, trust zones, procedural memory, local LLM routing, versioning, retention tiers, disclosure controls, project scoping, importers, benchmarks, recall X-ray, and more. These pages explain focused memory tracks inside that broader system.
What changes

The product idea in plain terms.

Capture the arc

A trace stores problem, investigation, root cause, and resolution as one chain-structured memory.

Route how-to queries

Questions like “how did we fix this last time?” can bias recall toward traces.

Keep detail intact

The intermediate steps stay visible instead of being compressed into a thin summary.

Already in Remnic
  • Decisions, facts, preferences, and procedures already have distinct categories.
  • Procedural memory is enabled by default after the #567 track.
  • Recall X-ray and explain surfaces help inspect why memories were retrieved.
Landed in this track
  • PR #575 added the reasoning_trace category.
  • PR #617 taught extraction to recognize reasoning chains.
  • PR #625 added retrieval boost and a dedicated subtree; PR #630 added the benchmark.
References
Related Remnic pages