Skip to content

YantrikDB

v0.9.3 Feature

This release adds 3 notable features for engineering teams evaluating rollout.

Published 13d AI Agents & Assistants
✓ No known CVEs patched
Read the diff → Tool health → What is this tool? →

✓ No known CVEs patched in this version

Topics

agent-memory ai-agents anthropic claude-code cognitive-memory database
+12 more
embeddings hnsw knowledge-graph llm llm-memory mcp memory persistent-memory python rust semantic-memory vector-db

ReleasePort's take

Moderate signal
editorial:auto 12d

The release adds several bugfixes and performance improvements while introducing a breaking change to `recall_demand` handling.

Why it matters: Bugfix severity 40 for validation contracts; breaking change severity 80 disables `recall_demand` on encrypted databases, requiring migration. Performance gains: distance‑path comparisons improve 7.39× (818 ns → 110 ns) and recall p95 latency halves (4–9.5 ms → 2.6–4.8 ms).

Summary

AI summary

Updates Correctness, Performance — measured, and Accuracy across a mixed release.

Changes in this release

Breaking High

`recall_demand` isolation is namespace‑keyed and disabled on encrypted databases, purging legacy rows via migration.

`recall_demand` isolation is namespace‑keyed and disabled on encrypted databases, purging legacy rows via migration.

Source: llm_adapter@2026-07-14

Confidence: high

 —
Performance High

Distance‑path comparisons are 7.39× faster (818.4 ns → 110.8 ns) via precomputed norms and SIMD AVX2+FMA kernels.

Distance‑path comparisons are 7.39× faster (818.4 ns → 110.8 ns) via precomputed norms and SIMD AVX2+FMA kernels.

Source: llm_adapter@2026-07-14

Confidence: high

 —
Performance Medium

End‑to‑end recall p95 latency roughly halved (4.0–9.5 ms → 2.6–4.8 ms).

End‑to‑end recall p95 latency roughly halved (4.0–9.5 ms → 2.6–4.8 ms).

Source: llm_adapter@2026-07-14

Confidence: high

 —
Bugfix Medium

`correct(new_text=...)` now returns `CorrectionRequiresReembed` (HTTP 422) until vector‑coherent correction path is implemented.

`correct(new_text=...)` now returns `CorrectionRequiresReembed` (HTTP 422) until vector‑coherent correction path is implemented.

Source: llm_adapter@2026-07-14

Confidence: high

 —
Bugfix Medium

Central numeric/vector contract gate validates inputs before side effects, returning typed errors instead of panics.

Central numeric/vector contract gate validates inputs before side effects, returning typed errors instead of panics.

Source: llm_adapter@2026-07-14

Confidence: low

 —
Bugfix Low

Keyword‑lane stopword hardening removes function‑word anchoring without regression on golden queries.

Keyword‑lane stopword hardening removes function‑word anchoring without regression on golden queries.

Source: llm_adapter@2026-07-14

Confidence: high

 —
Full changelog

v0.9.3 — correctness + performance train

Items 1–3 of the converged improvement plan (designed in a two-round debate with gpt-5.6-sol, grounded in code audit), plus a measured speed/accuracy program.

Correctness

  • Central numeric/vector contract gate: typed InvalidEmbedding { path, index, reason } / InvalidScalar { path, field, value } on every entry path — record, record_batch (whole-batch prevalidation), record_text (including the embedder''s own output — catches external-embedder NaN like the #60 ONNX 0/0 class), record_with_rid, insert_vector, embed, recall — validated before any side effect. Wrong-dimension insert_vector was a panic; now a typed error.
  • recall_demand isolation repair (schema v33): the v0.9.0 demand table was globally keyed and stored raw query text in plaintext even on encrypted databases. Now namespace-keyed (unscopable legacy rows purged by migration), and demand capture is fully disabled on encrypted databases. knowledge_gaps(namespace=...) is scope-explicit; session_digest(namespace=...) scopes decisions + conflicts for multi-tenant hosts.
  • Text corrections refused: correct(new_text=...) paired new text with the old embedding — corrected memories kept being retrieved under their old meaning. Now a typed CorrectionRequiresReembed (HTTP 422 with the workaround) until the vector-coherent correction path ships in v0.10. Metadata/importance/valence corrections unchanged; replication replay unaffected.

Performance — measured

  • 7.39× on the distance path (818.4 → 110.8 ns/comparison, dim 384, release): stored-vector norms precomputed at insert (were recomputed on every comparison), plus SIMD dot kernels — AVX2+FMA via runtime CPU detection (one wheel exploits whatever machine it lands on) with a portable ILP-unrolled fallback (aarch64 verified on CI). Kernels pinned to the sequential reference within 1e-9. Reproduce: cargo test --release -p yantrikdb --lib kernel_timing -- --nocapture --ignored.
  • End-to-end: scaling-benchmark recall p95 roughly halved (4.0–9.5ms → 2.6–4.8ms).

Accuracy

  • Keyword-lane stopword hardening (function words like "during" were anchoring keyword_match boosts on unrelated memories). No regression on the golden-query suite; the deeper ranking work (IDF-weighted keyword boosts, composite-score rebalance) is diagnosed with measurements and scheduled for v0.10.

Published to PyPI (pip install -U yantrikdb) and crates.io.

View diff on GitHub

Weekly OSS security release digest.

The CVE patches and breaking changes that affected production tools this week. One email, every Sunday.

No spam, unsubscribe anytime.

Share this release

Share on X Share on Bluesky

Track YantrikDB

Get notified when new releases ship.

Sign up free

About YantrikDB

All releases →

Related context

Earlier breaking changes

  • v0.7.20 `correct()` now mutates in place, preserving rid and adding revision history (BREAKING CHANGE).
  • v0.7.9 Pure-additive; existing engines keep English models on v0.1.0.

Beta — feedback welcome: [email protected]