v1.12.0

Added

• Added Hermes Agent as a first-class gno skill install --target hermes
  target, including --target all, uninstall, path inspection, and connector
  status support.

v1.11.0

Added

• Added packaged agent recipes for brain-first lookup, capture, meeting
  ingestion, email context, source summaries, idea capture, and citation
  discipline, with recursive gno skill show support and updated docs.

v1.10.1

Fixed

• Added a directory-lock fallback for platforms without lockf or flock, so
  capture writes and MCP write tools work on Windows.

v1.10.0

Added

• Added typed doc_edges with frontmatter relations: and active
  contentTypes[].graphHints projection metadata.
• Added bounded typed graph traversal through gno graph query, REST
  /api/graph/query, and MCP gno_graph_query.
• Added targeted retrieval diagnostics through gno query diagnose --target,
  REST /api/query/diagnose, and MCP gno_query_diagnose.
• Added --edge-type / --relation filters to gno links and
  gno backlinks for semantic relationship queries.

v1.9.0

Added

• Added second-brain note presets for original ideas, people,
  company/projects, and meetings, using flat preset frontmatter and a
  synthesis/timeline page pattern across capture surfaces.
• Added optional contentTypes config for schema-lite second-brain typing,
  warning-based rule normalization, and canonical contentType exposure in
  JSON search/query results.
• Updated repo docs, hosted docs, and bundled agent skill references for typed
  presets, synthesis/timeline pages, and content-type search metadata.

v1.8.0

Added

• Added gno capture for second-brain note capture from inline text, stdin, or
  files, with provenance frontmatter, tags, presets, JSON receipts, and quiet URI
  output.
• Added shared capture support across REST /api/capture, SDK
  client.capture(), MCP gno_capture, and Web UI Quick Capture, including
  parity tests and structured capture receipt schemas.

Changed

• Hardened capture writes with shared runtime validation, text-only input checks,
  disk-aware collision planning, exclusive create semantics, and MCP-only legacy
  overwrite handling.
• Updated GNO docs, specs, website sync, and bundled agent skill references for
  capture provenance behavior.

v1.7.1

Changed

• Made query-time graph expansion opt-in with gno query --graph and
  graph: true for API/MCP callers, keeping default hybrid queries on the
  faster BM25/vector/rerank path.

v1.7.0

Added

• Added gno doctor embedding fingerprint diagnostics for current freshness
  fingerprint, pending/stale chunks, legacy vectors, and mixed stored
  fingerprint groups.

v1.6.0

Changed

• Improved CPU embedding throughput defaults by using at most two native
  embedding contexts automatically, keeping low-memory Windows systems on a
  single context, and preserving GNO_EMBED_CONTEXTS=4 as an explicit
  benchmark-driven override.

Added

• Added a CPU embedding autoresearch benchmark script for comparing context pool
  variants with synthetic scheduler checks or real cached/downloaded GGUF
  embedding runs, and validating the Windows memory heuristic.
• Added a native embedding batch probe that verifies whether node-llama-cpp can
  retrieve distinct embeddings from multi-sequence native batches.

v1.5.2

Fixed

• Hardened local model backend startup on Windows by making node-llama-cpp
  source builds opt-in, timing out backend initialization, retrying CPU when
  automatic backend selection fails, and capping CPU embedding contexts on
  low-memory Windows machines.