Skip to content

v2.4.8 — RRF HNSW fallback fix + agent skills

Choose a tag to compare

View all tags
@isc-tdyar isc-tdyar released this 19 Jul 13:32
· 436 commits to main since this release
v2.4.8
db8b579

What's fixed

kg_RRF_FUSE — HNSW fallback (_engine/vector.py)

On Community Edition (or any deployment without an IVF index built), kg_RRF_FUSE
returned [] unconditionally. The index-registry loop only matched "ivf" type — the
"hnsw" type registered at engine init was never checked, so both vec_results and
txt_results stayed empty.

Now: when the registry contains "hnsw" but no "ivf", the vector leg falls through
to kg_KNN_VEC. HNSW + BM25 fusion works correctly on Community Edition.

ivf_build schema prefix (_engine/vector.py)

ivf_build hardcoded Graph_KG.kg_NodeEmbeddings in both SELECT paths, silently
ignoring set_schema_prefix(). Fixed to use _table("kg_NodeEmbeddings") consistently.

New tests

  • TestRRFFuseCommunityE2E (3 tests, ivg-iris) — regression pins: HNSW-only registry returns results, self-retrieval, HNSW+BM25 fusion
  • TestRRFFuseEnterpriseE2E (4 tests, ivg-iris-enterprise) — full IVF+BM25 RRF path, schema prefix round-trip, BM25 search
  • Unit tests — test_fuse_hnsw_uses_kg_knn_vec, test_fuse_hnsw_only_returns_results, test_ivf_build_uses_schema_prefix

Agent / developer experience

  • skills/iris-vector-graph/SKILL.md — agent skill file: Python API quickstart, key globals (^KG("tout"/"tin"/"tagg")), container setup, common gotchas (reserved words, timestamp format, schema prefix)
  • skills/ivg-arno/SKILL.md — Arno acceleration skill: when it matters, enable steps, ASQ vs Cypher, fixture patterns, TCP deploy
  • README: new AI Agent Development section
  • pyproject.toml: pip install iris-vector-graph[ai] installs iris-agentic-dev

Upgrade

pip install --upgrade iris-vector-graph==2.4.8

No API changes. No schema changes. Drop-in upgrade from any 2.4.x.

Assets 4
Loading