Source Intelligence

DisclaimerUnofficial, and not affiliated with Anthropic. Nearly all of this is read straight out of what ships: npm bundles, captured prompts, published docs. Anthropic's own notes go in verbatim, marked as theirs. The rest is my reading, and every entry carries the strings behind it. If one looks wrong, vote it down and say why.

All of v2.1.94 Home All releases olderv2.1.92 v2.1.96newer
Claude Code v2.1.94

LLM-Based Memory Synthesis

Feature flag
tengu_billiard_aviary Off in both readings

The flag server returned off for the account this site reads and for the anonymous baseline. A reading of off cannot rule out a rollout these two readings sit outside of.

This account: off · anonymous baseline: off · compiled default in v2.1.94: no gate table built for this version

These values were read against a different version of Claude Code, so treat them as the nearest reading available instead of one taken on this release.

Read once, for one account on one subscription tier, against v2.1.94. It isn't a statement about your account. What a flag value here can and cannot tell you

What

New memory retrieval path that uses an LLM to synthesize relevant memories into a single paragraph, replacing the previous raw-file retrieval approach.

Details
  • When tiny memory mode (tengu_billiard_aviary) is active, uses LLM synthesis to produce contextually relevant memory summaries
  • System prompt instructs: "synthesize information directly relevant to" the current query
  • Output formatted as JSON with a one-paragraph synthesis
  • Recalled memories shown to users with: "Recalled from your persistent memory system:"
  • Synthesis memories are excluded from the standard memory context to avoid duplication
Evidence

Memory synthesis (search for "Synthesize memory information relevant to", "synthesizeRelevantMemories")

Strings lifted out of the shipped bundle, so the claim above can be checked against them.

See this entry in the whole of v2.1.94 →