LearnNewsExamplesServices
Frontmatter
id12700
title[v13 release-notes] Memory Core + telepathy chapter
stateClosed
labels
documentationai
assigneesneo-opus-vega
createdAtJun 7, 2026, 11:01 PM
updatedAtJun 29, 2026, 4:26 PM
githubUrlhttps://github.com/neomjs/neo/issues/12700
authorneo-opus-vega
commentsCount1
parentIssue12696
subIssues[]
subIssuesCompleted0
subIssuesTotal0
contentTrust
projected
quarantined1
signals[]
blockedBy[]
blocking[]
closedAtJun 8, 2026, 2:12 AM

[v13 release-notes] Memory Core + telepathy chapter

neo-opus-vega
neo-opus-vega commented on Jun 7, 2026, 11:01 PM

Sub of #12696 (v13 release). The Memory Core chapter — the #1 ai/ change-mass in v13 (33 MC files / ~16.7k insertions) and a hero area. Verified detail lives in the #12695 Preview comments (batches 1–4).

What shipped (verified)

  • Chronological recency recallquery_recent_turns (#12672): the recency retrieval axis the MC lacked (it had only semantic).
  • miniSummary backfill (#12676 / #12673): backfill of the pre-existing memory corpus.
  • REM / Dream / Golden Path digestion + degraded-query surfacing (missing/corrupt retrieval becomes visible, not a silent context loss).
  • The Native Edge Graph (SQLite, new in v13) as the MC's structural layer.

The differentiating capability: telepathy (read-only cross-agent introspection)

The MC's multi-agent model is per-author-owned, append-only memories + read-only, trust-tiered cross-agent introspection ("telepathy"): an agent reads a peer's memories — gated by the 8-tier trust taxonomy + AUTHORED_BY provenance + stable AgentIdentity — and never overwrites them. This is the swarm's transparent introspection (the README's "read each other's reasoning through shared memory"), and the mechanism the session-sunset emblem demonstrates.

  • Emblem: the Hallucinated Sunset Protocol (#10370) — a Claude's hallucinated reasoning, read by Gemini through the MC, became a governed skill.

Acceptance Criteria

  • The v13 release-notes Memory Core chapter (in the #12695 artifact) covers the shipped capabilities above + telepathy framed as a capability (transparent cross-agent introspection), with verified anchors.
  • Public wording vetted before publish; no forward roadmap / strategic-positioning material in the public notes.

Refs #12696

tobiu referenced in commit be35b3c - "docs(release): add v13 telepathy + context-recovery section (#12700) (#12709)" on Jun 8, 2026, 2:12 AM
tobiu closed this issue on Jun 8, 2026, 2:12 AM
edihasaj
edihasaj Jun 29, 2026, 4:26 PM

Nice work on the Memory Core and the telepathy capability, that append-only provenance and AgentIdentity model are exactly the kinds of primitives that make memory portable. If you are interested in making those memories interoperable with other agent frameworks and vector stores, I maintain an open spec and adapter-focused repo called Universal Memory Protocol that defines schemas, identifiers, provenance metadata, and sync semantics you could map to Neo's MC. Would you be open to a short discussion or a PR to prototype an adapter so Neo memories can be exported/imported by other tools? [QUARANTINED_URL: github.com]