[v1.1.0] - Trendsetter - 2026-05-12

NornicDB v1.1.0 Release Notes

⚠️ ⚠️ ⚠️ READ BEFORE UPGRADING — STORAGE FORMAT CHANGES ⚠️ ⚠️ ⚠️

This release ships breaking storage changes that require a one-time
in-place migration on first start. Once your data files are upgraded,
older versions of NornicDB will not be able to read them.

🦀 Back up your data directory before upgrading. 🦀

Run the database with the --upgrade-storage flag (or, in the macOS
installer, tick the upgrade checkbox on the first-run wizard, or in the
tray settings panel restart with the flag enabled). This is a rolling
upgrade — when it completes, the engine compacts the on-disk layout to
the new format and resumes normal operation.

After this release, no further major storage overhauls are planned. 🤟

Release tag: v1.1.0
Range covered: v1.0.45..v1.1.0
Combined repository delta: 644 files changed, +170,459 / −9,816 lines across 91 non-merge commits.

This document is a single rolled-up changelog for everything shipped in v1.1.0
across both the v1.1.0-preview (2026-05-12) and v1.1.0-rc (2026-05-14) and v1.1.0 prime (2026-05-15)
milestones. Each section merges the entries from both tags so operators
upgrading directly from v1.0.x see the complete feature set.

Changes from RC:

• Knowledge-policy ON ACCESS recording scope:

  • moved ON ACCESS recording out of storage visibility checks to trigger only for entities actually materialized into query results.
  • routed recording through storage wrappers to preserve namespaced IDs correctly.
  • extended e2e coverage with positive and inverse test cases including accumulator contents before flush.

• MVCC commit ordering at sequence saturation:

  • hardened MVCC commit ordering when the global commit sequence reaches MaxUint64.
  • The reason for the monotonic counter is that we can actually record transactions so fast that slower processors can have negative nanos drift causing random conflicts with serial ingesiton. @1 million commits/sec it would exhaust in 584.5 million years. @1 billion/sec - 584,500 years.
  • keep sequence pinned and fall back to strictly increasing high-water timestamps rather than wrapping or failing in said years.
  • updated snapshot conflict detection to use timestamp ordering only for the saturated equal-sequence case.
  • added focused regression tests for the saturation fallback.

Added

Knowledge-policy administration via Cypher

• CREATE, ALTER, DROP, and SHOW support for knowledge-policy
  management — decay profiles and promotion policies are administered
  directly through Cypher DDL.
• Built-in CALL nornicdb.knowledgepolicy.* procedures for inspecting
  policy state, profiles, resolution, and deindexing status.
• Browser-side management UI now drives the Cypher-backed surface instead
  of a separate admin API.

Knowledge-policy runtime and lifecycle scoring

• New shared-profile resolver, access-metadata indexing,
  scoring-before-visibility, and suppression/deindex infrastructure.
• Cypher functions for policy-aware scoring and diagnostics:
  decayScore, decay, and policy.
• Inverted Ebbinghaus-style decay curves — set a negative
  halfLifeSeconds to flip any decay function family in place. Combined
  with the LAST_ACCESSED anchor, this implements Roynard-style
  consolidation where idle time strengthens the score and access resets
  it (see docs/user-guides/decay-profiles.md).

Hybrid-search property pre-filter

• POST /nornicdb/search now accepts an optional filters field that
  restricts the candidate set by property values before top-K
  vector/BM25 selection, preserving recall on sparse filtered sets.
• Filter semantics: OR within a key
  ({"collection": ["user_a", "user_b"]}), AND across keys
  ({"collection": ["user_a"], "type": ["text"]}); scalar and array
  property values both supported.
• Applied in rrfHybridSearch, vectorSearchOnly, and
  fullTextSearchOnly paths.
• Primary use case: multi-tenant RAG where each node carries a
  collection array identifying which users may access it.

End-to-end observability stack

• OpenTelemetry tracing across HTTP, Cypher planning and execution, Bolt
  sessions, storage operations, embeddings, and search flows.
• Helm chart support for observability deployment, including
  ServiceMonitor, PodMonitor, startup/readiness/liveness probes, and
  default network-policy wiring.
• Curated Grafana dashboard, Prometheus alert rules, and an
  auto-generated metrics reference covering the current exported metric
  catalog.
• pprof hooks for goroutine and mutex profiling.

Storage internals and observability

• Internal numeric IDs and a freelist-based key reclamation path to
  reduce storage overhead while keeping keys monotonic and reusable.
• Per-namespace property-key dictionary as part of the storage-v2 path.
• Migration progress and statistics logging during in-place upgrades.

Changed

Knowledge-policy operator surface

• Knowledge-policy administration moved off the dedicated HTTP API and
  onto the new Cypher DDL + procedure model.
• Admin UI, runtime behavior, and diagnostics aligned around the new
  suppression-anchor and decay-profile model.

Storage serialization

• Removed Gob as an active serializer while preserving forward migration
  support for older data written with Gob.
• Storage-v2 hot path emits the tokenized property-key codec
  exclusively; the v1 codec is read only by the in-place migration.

Write-heavy query execution

• Bulk-insert and canonical write paths expanded and tuned so more
  high-volume ingestion shapes avoid unnecessary reads and stay on the
  optimized execution path.
• Refreshed bulk-insert cookbook and related operator guidance.

UI

• Browser management pages updated to use newer uiGrid capabilities
  for database and retention administration views.

Bundled runtimes

• Upgraded the bundled llama.cpp integration to b9106.

Fixed

Cypher correctness

• NOT IN parsing — predicate now correctly negates the IN membership
  check.
• Knowledge-policy ON ACCESS arithmetic coercion — runtime type
  coercion now matches Cypher DDL expectations (no silent truncation of
  floats by integer-typed parameters).

Transaction conflict handling

• Adjusted timing in conflict-sensitive paths to reduce avoidable write
  conflicts.
• Snapshot-isolation conflict detection now uses commit-sequence
  comparison instead of full timestamp comparison, eliminating false
  positives on hosts with non-monotonic wall clocks (CI flake fix).
• Unique-merge conflict handling fixed so retryable conflict errors are
  surfaced and scoped correctly for Bolt clients.

Storage migration and recovery

• Storage migration now creates all expected indexes during the in-place
  upgrade (label, outgoing, incoming, edge-type), eliminating the
  "edges disappeared after migration" issue.
• MVCC schema metadata recovery fixed.
• Edge-between deindexing behavior fixed.
• Schema lock leak scenarios resolved.
• Tightened uniqueness-vs-constraint-contract handling in the affected
  storage and Cypher paths.

Hybrid search filters

• Empty filter-list handling fixed so requests with empty filter lists
  no longer discard all results.

Knowledge-policy correctness

• Decay diagnostics, targeted suppression invalidation, score-argument
  handling, and reveal-path correctness fixed.
• Embed and reveal worker concurrency hardened for the new knowledge
  lifecycle features under load.

Lifecycle and high-watermark timing

• Lifecycle and high-watermark nanosecond timing bugs affecting
  fast-write scenarios fixed.

Observability safety

• Trace redaction and baggage filtering fixed so credentials and other
  sensitive values are not emitted into spans.
• Nil-pointer and span-property edge cases in the observability pipeline
  fixed.

Cross-platform build and packaging

• Windows and macOS build and packaging regressions affecting release CI
  and local packaging flows fixed.

Compatibility

MVCC is now opt-in

• The default retained-version count is now 0, which disables MVCC
  history unless explicitly enabled in configuration. Operators who
  rely on as-of reads or audit trails must opt back in by setting
  NORNICDB_MVCC_RETENTION_MAX_VERSIONS (or the equivalent config
  field).

Storage format

• Data written by v1.1.0 is not readable by v1.0.x. Plan the upgrade
  accordingly. The migration is one-way; rollback requires restoring
  from your pre-upgrade backup.

Documentation

• Added and refreshed:
  • Metrics reference generated from the current observability catalog.
  • Helm, dashboard, and alerting guidance for the new observability
    stack.
  • Knowledge-policy and decay diagnostics documentation aligned to the
    Cypher-backed administration model.
  • Memory-decay and decay-profile documentation for the new inverted
    decay behavior, including the scoreFloor vs visibilityThreshold
    disambiguation and lifecycle examples.
  • User guidance around promotion policy and Ebbinghaus/Roynard
    bootstrap behavior.
  • Bulk-insert cookbook material for the current optimized write paths.

Upgrade Checklist

1. Take a complete backup of your data directory. No exceptions —
   the migration is in-place and not reversible without a restore.
2. Stop the running NornicDB instance cleanly so the WAL flushes.
3. Start v1.1.0 with --upgrade-storage (CLI), the macOS installer
   checkbox, or the equivalent tray-settings flag.
4. Watch the migrati...

[v1.0.45] - Go - 2026-05-08

Added

• Broader hot-path coverage for UNWIND MATCH SET updates:
  • added a batched UNWIND + MATCH + SET hot path so more update-heavy Cypher write workloads can stay on the optimized execution path.

Changed

• Browser query-results grid behavior:

  • updated the web UI to @ornery/ui-grid-* 1.0.5, bringing the latest resize and scroll tracking behavior from the new grid runtime.
  • query-results selection now relies on the grid's per-column opt-out behavior for the synthetic Select column, so hidden sort and group controls follow the legacy ui-grid behavior instead of rendering as disabled header buttons.

• Cypher write-path routing:

  • tightened MATCH ... SET hot-path routing so optimized update execution only applies when uniqueness guarantees are present, reducing incorrect fast-path routing for ambiguous match shapes.

• Maintenance and dependency refreshes:

  • refreshed UI grid dependencies from the 1.0.4 line to 1.0.5 and regenerated the UI lockfile to match the published package graph.
  • updated package metadata and peer-dependency wiring in the UI package so installs resolve cleanly against the latest grid release.
  • folded in mainline merge-conflict cleanup that affected the release range.

Fixed

• Embedding worker lifecycle stability:

  • fixed overlapping embed-worker reset and close flows by serializing worker lifecycle transitions, preventing wait-group reuse panics during concurrent embedding workloads.

• Grid column opt-out chrome:

  • fixed the query-results Select column so column-level sorting and grouping opt-outs now hide header controls using the grid's native per-column behavior instead of relying on CSS workarounds.

Technical Details

• Range covered: v1.0.44..HEAD
• Commits in range: 9 (non-merge)
• Repository delta: 15 files changed, +1,322 / -582 lines
• Primary focus areas: query-results grid upgrades, UNWIND MATCH SET hot-path expansion and safety, embedding-worker lifecycle stability, and targeted UI package maintenance.

[v1.0.44] Lithium - 2026-05-02

Added

• Convenience defaults for vLLM-backed local generation:

  • added vLLM-friendly defaults in the Heimdall/OpenAI-compatible generation path so local and self-hosted inference setups require less manual configuration.

• Relationship-head recovery fallback for storage indexes:

  • added an edge-between head-index fallback path so relationship lookups can recover more reliably when derived index state needs repair or reconstruction.

• Named relationship-set support in batched merge flows:

  • added support for named relationship sets in UNWIND merge batches, expanding the structured relationship-merge shapes that can stay on the optimized execution path.

Changed

• Relationship merge and probe execution paths:

  • optimized relationship probe planning and batched SQL relationship writes so more merge-heavy Cypher workloads avoid redundant work.
  • shared node-lookup cache locking across executor and transaction clones to reduce duplicated lookup coordination during concurrent execution.

• Transient error classification and retry signaling:

  • refactored transient conflict handling into the shared pkg/errors surface and mapped Bolt transaction conflicts onto explicit transient error semantics.
  • improved server-side failure handling so retryable conflict conditions are surfaced more consistently to clients.

• Storage relationship-index maintenance:

  • hardened edge-between index repair, exact relationship lookups, and streamed edge index rebuild behavior to keep relationship indexes aligned with stored graph state.

• Documentation and dependency maintenance:

  • refreshed the README, retention-policy guide, and hot-path query cookbook to match the current relationship hot-path behavior and operator guidance.
  • updated Go module dependencies and UI package dependencies as part of the release range.

Fixed

• Relationship merge property-set correctness:

  • fixed relationship merge property updates so property sets are applied correctly across merge execution paths.

• Relationship hot-path edge cases:

  • fixed edge-case handling in relationship hot paths, including exact edge lookup behavior and fallback coverage for edge-between index state.

• Bolt transient failure behavior:

  • fixed Bolt failure handling so conflict-driven transaction failures are decoded and classified more robustly for clients.

Tests

• Added and expanded coverage for:
  • relationship merge property handling and named relationship-set merge batches
  • relationship hot-path routing, SQL relationship writes, and probe optimization behavior
  • Bolt transient error mapping and failure-chunk decoding
  • storage edge-between index repair, streamed rebuild behavior, and exact relationship lookup regressions
  • shared-state isolation in affected regression tests

Documentation

• Added and refreshed:
  • retention policy operator guidance
  • README release-facing behavior notes
  • hot-path cookbook coverage for the current relationship query optimizations

Technical Details

• Range covered: v1.0.43..HEAD
• Commits in range: 23 (non-merge)
• Repository delta: 40 files changed, +2,494 / -255 lines
• Primary focus areas: relationship merge correctness, relationship hot-path and edge-index reliability, transient Bolt conflict handling, convenience inference defaults, and targeted documentation/dependency refreshes.

[v1.0.43] Clockwork - 2026-04-24

Special thanks to @linuxdynasty for submitting a bunch of fixes and optimizations!

Added

• Runtime retention and privacy control plane:

  • added runtime retention sweep management in the database layer, including sweep budgeting, excluded-label handling, record collection, and immediate sweep triggering.
  • added admin retention endpoints for policies, legal holds, erasure requests, manual sweeps, and retention status.
  • added GDPR integration hooks so privacy and retention workflows can share the same server-side control surface.

• Retention administration UI:

  • added a dedicated Retention Admin page in the web UI with policy editing, legal hold management, erasure request processing, defaults loading, and sweep controls.
  • wired new client API types and routes so operators can manage retention behavior from the browser instead of only through raw admin calls.

• Constraint and planner regression coverage:

  • added extensive deterministic tests for constraint contract parsing, schema validation, Bolt database-manager behavior, transaction wrapper reuse, UNWIND merge batches, and fabric helper types.
  • added regression coverage around database-scoped rollback handling and deterministic multidatabase enforcement behavior.

• Knowledge-layer documentation and research material:

  • added a full knowledge-layer persistence implementation plan, expanded design material, Kalman-focused planning docs, and a new research-paper draft on persistence semantics. (v1.1.0)

Changed

• Retention configuration and operator surface:

  • expanded runtime config, example YAML, and macOS/app UI surfaces for retention and memory-management controls.
  • made chunk overlap configurable in the macOS app and web UI chunking flow.

• Cypher MERGE and UNWIND MATCH routing:

  • routed more UNWIND + MATCH/map-merge shapes through the hot path and tightened optional-match fallback behavior.
  • started using unique-constraint and unique-index information more aggressively for merge lookup and validation paths.

• Constraint contract parsing and validation internals:

  • replaced older regex-heavy constraint-contract parsing with allocation-free keyword scanning helpers.
  • normalized unique-constraint cache keys and extended contract matching to cover broader label combinations and namespace cases.

• Documentation and release content:

  • refreshed README, operations docs, compliance docs, user-guide indexes, and the hot-path cookbook to match the current runtime behavior.
  • removed or folded obsolete planning material into the newer knowledge-layer and compliance documentation set.

• Dependency maintenance:

  • refreshed Go module dependencies and updated UI package dependencies as part of the release range.

Fixed

• Database-scoped Bolt rollback correctness:

  • fixed Bolt explicit transaction rollback so rollback is honored against the correct selected database instead of leaking across database scope boundaries.
  • hardened transaction wrapper reuse paths and related rollback regressions in Bolt/Cypher integration tests.

• Constraint validation correctness and safety:

  • fixed unique lookup and validation paths to guard non-comparable values and avoid incorrect equality handling during unique checks.
  • corrected constraint-validation plumbing so unique indexes and schema checks stay aligned across storage and Cypher surfaces.

• Local LLM generation stability:

  • fixed CGO generation-path issues in the local llama integration, including causal-attention setup and health checks for non-unified KV caches.

• Deterministic multidatabase enforcement tests:

  • fixed rate-limit and enforcement test behavior to avoid timing-driven flakes and keep multidatabase coverage deterministic.

Tests

• Added and expanded coverage for:
  • retention manager behavior, retention admin handlers, and privacy/erasure workflows
  • Bolt database-manager integration, database-scoped rollback, and transaction-wrapper reuse
  • UNWIND/MATCH/MERGE hot-path routing and fallback behavior
  • constraint contract parsing, validation, schema interactions, and unique lookup edge cases
  • fabric helper coverage, plan-cache behavior, and multidatabase enforcement determinism

Documentation

• Added and refreshed:
  • GDPR compliance guidance and retention-policy operator documentation
  • knowledge-layer persistence plan and implementation details
  • Kalman behavioral signal planning notes
  • persistence-semantics research paper draft, evaluation plan, bibliography, and related-work notes

Technical Details

• Range covered: v1.0.42-hotfix..HEAD
• Commits in range: 47 (non-merge)
• Repository delta: 64 files changed, +11,730 / -1,528 lines
• Primary focus areas: retention/GDPR operator workflows, Bolt rollback correctness, unique-constraint driven planner and validation improvements, deterministic regression coverage, and major knowledge-layer documentation expansion.

[v1.0.42-hotfix] Anthem Pt. 2 - 2026-04-17

Added

• Explicit Bolt host binding support:

  • added a dedicated Bolt host field so the Bolt listener can bind to the resolved server address instead of always falling back to the wildcard socket.

• Bind-address regression coverage:

  • added CLI/config/environment precedence tests for server bind resolution.
  • added Bolt listener tests to verify loopback binding and secure default host behavior.

Changed

• Bind-address resolution precedence:

  • centralized bind-address resolution for serve startup so CLI flags, loaded config, and explicit protocol-specific environment variables are resolved consistently before HTTP and Bolt are started.
  • aligned startup endpoint display output with the resolved address, while still presenting localhost for wildcard binds in user-facing logs.

• Dependency maintenance:

  • updated github.com/hybridgroup/yzma to v1.12.0.
  • refreshed related Go module entries pulled in by the dependency update.
  • updated UI dev dependencies: postcss to 8.5.10 and typescript to 6.0.3.

Fixed

• Bolt listener exposure:
  • fixed the Bolt server so configured bind addresses now apply to the actual listener instead of only affecting user-facing startup messages.
  • restored secure loopback binding as the default when no explicit override is provided.

Tests

• Added and expanded coverage for:
  • CLI --address precedence over config and environment sources
  • config-file and explicit environment bind-address overrides
  • Bolt listener binding to the configured host instead of wildcard interfaces

Technical Details

• Commit covered: adce4f9a9fc7b6aada07c0bfa2d737cd7a6efaca
• Commits in release: 1
• Repository delta: 8 files changed, +194 / -26 lines
• Primary focus areas: Bolt bind-address correctness, secure default listener behavior, regression coverage, and targeted Go/UI dependency maintenance.

[v1.0.42] Anthem - 2026-04-16

The answer to life the universe and everything...

Anthem allows you to utilize the underlying model inside NornicDB with Continue VSCode Plugin or any other OpenAI-compatible chat interface.

Added

• Heimdall tool coverage and documentation:

  • exposed local chat tool routing more consistently so the OpenAI compatible endpoint can surface the user's tools from an agent into injected actions in chat flows.
  • added and refreshed Heimdall, knowledge-layer, and compliance documentation, including new guides for the Heimdall AI assistant and knowledge-layer persistence planning.

• Regression coverage for transactional and constraint edge cases:

  • added regression coverage for constraint handling in Bolt/Cypher paths.
  • added targeted tests for async flush race conditions and complex UNWIND/MERGE fallback behavior.

Changed

• Heimdall local chat behavior:

  • hardened action parsing, tool forwarding, and local chat prompt behavior so action envelopes are handled more predictably.
  • simplified the watcher hello/search tool flow and aligned injected action naming with the current plugin surface.

• Cypher hot paths and fallback routing:

  • generalized hot-path execution to support broader n-ary and generic query shapes.
  • refined UNWIND + MERGE fallback routing so complex query shapes are handed off to the full executor more consistently.
  • preserved transaction-visible lookups in fallback execution paths to keep behavior aligned with in-transaction state.

• Documentation and release content:

  • folded older planning material into the current docs set and refreshed migration, AI-agent, and deployment-facing documentation.

Fixed

• MERGE cache and transaction correctness:

  • fixed stale MERGE cache invalidation after implicit transaction aborts.
  • fixed async flush timing so MVCC heads do not advance while a transaction still depends on its visible state.

• Installer and model download reliability:

  • updated installer/download locations and added validation so failed model downloads are not accepted as valid artifacts.
  • improved macOS model download handling to verify downloaded files before treating them as usable GGUF models.

• Heimdall local chat stability:

  • fixed local chat template replay and unknown tool-call forwarding behavior.
  • hardened Heimdall test races and related local chat execution edge cases.

Tests

• Added and expanded coverage for:
  • async engine flush/count race handling
  • UNWIND/MERGE fallback and optional lookup regression paths
  • constraint regressions across Bolt and Cypher execution
  • Heimdall local chat tool execution and race-sensitive paths

Technical Details

• Range covered: v1.0.41..HEAD
• Commits in range: 19 (non-merge)
• Repository delta: 48 files changed, +5,437 / -3,007 lines
• Primary focus areas: Cypher fallback correctness, transaction visibility and async flush safety, Heimdall local chat/tool handling, installer download validation, and documentation consolidation.

v1.0.41 Blue Orchid

Blue Orchid brings us a simple graph explorer, fast-path enhancements, and other vairous fixes for the canonical graph ledger setup.

I'm also playing around with a side-project that I'm using to try and test out the temporal functions and graph explorer by reconstructing the code state of a given git repository based on commit hash. Git-to-Graph

Added

• Graph explorer UI:

  • added a new graph explorer tab with neighborhood visualization and node-details integration.
  • wired the browser page, node details panel, and client API helpers to support the new exploration flow.

• MERGE and string-literal infrastructure:

  • added shared string-literal decoding helpers so Cypher literal handling is consistent across execution paths.
  • expanded merge-chain and relationship-shape support for the fallback and hot-path planner routes.

Changed

• UNWIND and MERGE hot paths:

  • generalized the staged compound UNWIND mutation path and tightened the merge-chain hot path so parameter handling and optional lookups behave consistently.
  • memoized the merge-chain path where the executor can safely reuse the structured shape detection.

• Cypher executor plumbing:

  • updated executor, pattern parsing, property evaluation, and clause handling to align the new hot-path routing logic.
  • refreshed the query-shape tests and regression coverage around compound UNWIND, merge chains, and literal decoding.

• Release and dependency maintenance:

  • updated README/version metadata and release workflows for the new build.
  • refreshed dependency pins across Go, UI, llama.cpp, Docker, and build scripts.
  • updated the hot-path performance cookbook to document the new query-shape coverage.

Fixed

• Fallback MERGE binding preservation:

  • fixed fallback MERGE chain handling so bindings are preserved correctly when the structured hot path is not selected.
  • restored parameter handling in the generalized UNWIND merge-chain route.

• Linker and packaging stability:

  • fixed the native linker path used by the CUDA build flow.
  • cleaned up version bump artifacts so the packaged build reports 1.0.41 consistently.

Tests

• Added and expanded coverage for:
  • compound UNWIND mutation staging and merge-chain hot-path routing
  • fallback MERGE chain bindings and optional lookup behavior
  • string-literal decoding and property evaluation helpers
  • graph explorer browser integration and node-details interactions
  • executor trace coverage and query-shape regression cases

Technical Details

• Range covered: v1.0.40..HEAD
• Commits in range: 10 (non-merge)
• Repository delta: 50 files changed, +3,163 / -502 lines
• Primary focus areas: Cypher merge/UNWIND hot-path generalization, literal decoding cleanup, graph explorer UI delivery, dependency refreshes, and packaging/build fixes.

v1.0.40 Kiyote

Everyone, thank you all! We crossed 500 stars in 3 months! Thank you all for coming along with me on my journey 🫶

Kiyote is all about optimization, reliability and hardening along the hot path. We got rid of the old regexes that were sitting around and replaced them with keyword scanning, which is much more efficient. (see below)

Parser Validation Benchmark: Nornic vs ANTLR

Isolated parser-validation benchmark on Apple M2 Max (darwin/arm64) using:

go test ./pkg/cypher -run '^$' -bench BenchmarkParserValidationIsolation -benchmem -count=1

Speedup is ANTLR ns/op ÷ Nornic ns/op.

Headline Numbers

• Nornic was faster on every isolated parser-validation shape measured.
• Speedups ranged from 50.8x to 173.4x.
• Nornic stayed at 0 allocs/op across the full suite.
• ANTLR ranged from 30 to 149 allocs/op.

Code-wide, the advantage comes from four things:

Nornic uses specialized string scanners instead of a full grammar pipeline. In pkg/cypher/executor.go, validateSyntaxNornic is mostly bounded checks over raw bytes: valid starting clause, balanced delimiters, and lightweight structural validation. That keeps the hot path in straight-line code with predictable branching.

Nornic avoids lexer, token stream, parse tree, and grammar machinery entirely on its fast path. ANTLR validation in pkg/cypher/antlr/parse.go still has to run a lexer, build a token stream, drive the parser automaton, and potentially retry from SLL to LL. Even pooled, that is fundamentally more work.

Nornic is heavily optimized around query-shape helpers and direct scans. Files like pkg/cypher/string_patterns.go, pkg/cypher/query_patterns.go, pkg/cypher/traversal.go, and pkg/cypher/compound_query_shape_matcher.go are written to recognize exactly the query structures the engine cares about, without paying for general-purpose parse-tree construction.

Nornic stays allocation-free on the isolated validation path, while ANTLR still allocates heavily. That is the visible benchmark result, but it is really a symptom of the design: Nornic validates directly against the input string; ANTLR builds intermediate parser state objects because it is solving a more general parsing problem.

TLDR; Nornic is faster because it is a purpose-built, zero-allocation, scanner-driven validator/executor front end, while ANTLR is a general grammar engine with lexer/token/parser overhead and fallback complexity. The speedup is mostly architectural, not just micro-optimization.

[v1.0.40] "Kiyote" - 2026-04-11

Added

• Structured Cypher hot-path matchers:

  • replaced several regex-based fast-path routers with scanner/helper-based matchers for compound queries, UNWIND routes, traversal shapes, shortest-path handling, and related query-shape detection.
  • added explicit hot-path trace coverage so the executor can report when the new structured matchers are used.

• Parser comparison reporting:

  • added an integrated Nornic-vs-ANTLR comparison harness that prints per-query timing ratios and an overall summary.
  • stabilized the comparison output so users and maintainers can trust the speedup numbers instead of single-run noise.

Changed

• Traversal and match planning:

  • improved traversal planning for top-k seeded lookups so multi-key ORDER BY shapes can still use the indexed path when possible.
  • compiled common binding-row WHERE predicates and made traversal execution honor the active temporal viewport consistently.
  • refined MERGE routing for composite-key patterns and hardened batch deduplication so key collisions are handled predictably.

• Cypher execution cleanup:

  • migrated hot-path query-shape parsing away from regex capture arrays and onto shared helpers.
  • removed obsolete hot-path regex definitions after the structured matcher cutover.
  • refreshed parser-mode and query-pattern plumbing so the same user-facing queries route more consistently across executor paths.

• Documentation and release hygiene:

  • refreshed README and deployment notes to reflect the current product direction.
  • updated performance notes for hybrid Bolt queries.
  • refreshed dependencies as part of the release cycle.

Fixed

• Shortest-path parsing regression:

  • fixed a panic in single-MATCH shortest-path queries by correcting the clause boundary scan used to recover the preceding MATCH clause.
  • added regression coverage for the exact Bolt fallback query shape that exposed the issue.

• Query-shape stability:

  • restored deterministic ordering and route selection for the query families that were being optimized in the hot path.
  • tightened parser and traversal tests around the new helper-based implementations.

Tests

• Added and expanded coverage for:
  • structured compound-query matcher behavior and reject reasons
  • hot-path trace assertions for compound query routing and traversal seeding
  • shortest-path parsing, execution, and regression handling
  • UNWIND collect/distinct and batch merge helper paths
  • parser comparison output and performance reporting

Technical Details

• Range covered: v1.0.39..HEAD
• Commits in range: 13 (non-merge)
• Primary focus areas: Cypher hot-path matcher migration, traversal and merge routing, parser comparison tooling, shortest-path correctness, and release documentation updates.

v1.0.39 - Punkrocker

Punkrocker is the latest release of NornicDB, focused on stronger graph data contracts, more predictable transaction behavior, better bulk-write stability, and faster traversal execution for real-world query shapes.

This release adds a new way to define grouped schema rules, expands relationship policy enforcement, improves parser compatibility, and tightens correctness in explicit transactions and ordered traversal queries.

Highlights

Block-style constraint contracts

You can now define multiple related rules together in a single constraint contract using REQUIRE { ... }.

This makes it easier to express real data rules in one place instead of scattering them across separate constraint definitions.

CREATE CONSTRAINT person_contract
FOR (n:Person)
REQUIRE {
  n.id IS UNIQUE
  n.name IS NOT NULL
  n.status IN ['active', 'inactive']
  (n.tenant, n.externalId) IS NODE KEY
  COUNT { (n)-[:PRIMARY_EMPLOYER]->(:Company) } <= 1
}

Relationship contracts work too:

CREATE CONSTRAINT works_at_contract
FOR ()-[r:WORKS_AT]-()
REQUIRE {
  r.id IS UNIQUE
  r.startedAt IS NOT NULL
  startNode(r) <> endNode(r)
  startNode(r).tenant = endNode(r).tenant
  r.hoursPerWeek > 0
}

You can inspect these grouped contracts separately with:

SHOW CONSTRAINT CONTRACTS

Relationship cardinality constraints

Punkrocker adds directional relationship count limits, making it easier to enforce edge-level policy directly in the schema.

CREATE CONSTRAINT max_employers
FOR ()-[r:WORKS_AT]->()
REQUIRE MAX COUNT 3

This is useful for graph rules like:

• limiting how many active outbound edges of a type a node can have
• enforcing single-primary or bounded-assignment patterns
• keeping relationship structure clean without application-side checks

Better parser compatibility

This release broadens Cypher compatibility around newer REQUIRE forms and improves support for WITH EMBEDDING and related query shapes.

That means more valid user queries work consistently across parser modes, including richer constraint syntax and embedding-enabled write flows.

Example:

CREATE (n:Doc {id:'d1', content:'hello world'})
WITH EMBEDDING
RETURN count(n) AS created

Explicit transaction correctness

Punkrocker improves explicit transaction behavior so validation happens against the final committed state, not against temporary intermediate states inside a transaction.

This matters for multi-step writes where a later statement makes the data valid before commit.

Example transaction flow:

BEGIN
CREATE (a:Person {id:'p1'})
CREATE (b:Company {id:'c1'})
CREATE (a)-[:WORKS_AT {id:'w1', hoursPerWeek: 0}]->(b)
MATCH ()-[r:WORKS_AT {id:'w1'}]->() SET r.hoursPerWeek = 40
COMMIT

In Punkrocker, this kind of flow is validated correctly at commit time based on the final state.

More stable bulk writes

Large UNWIND and batch update pipelines are more robust in this release.

Punkrocker reduces write amplification when a row creates a node and then updates it again in the same transaction, helping avoid oversized transaction failures in heavy write workloads.

Example:

UNWIND $rows AS row
CREATE (n:Doc {id: row.id})
SET n = row
RETURN count(n)

Faster ordered traversal queries

This release adds a generic relationship ORDER BY ... LIMIT planner that can seed traversal from the filtered end-node side using index-backed top-k lookups.

In practice, that improves latency for ordered traversal queries without relying on hardcoded, application-specific rewrites.

Example:

MATCH (o:OriginalText)-[:TRANSLATES_TO]->(t:TranslatedText)
WHERE t.language = 'fr'
RETURN o, t, t.createdAt
ORDER BY t.createdAt DESC
LIMIT 10

More predictable ordering for graph query results

Punkrocker fixes ordering behavior for relationship-heavy query results, including:

• ordering by aliased expressions
• ordering by returned node properties
• cases where labels or relationship types such as :Order or :ORDERS previously interfered with ORDER BY detection

Formal Patch notes:

Added

• Block-style constraint contracts (REQUIRE { ... }):
  • added grouped schema contracts so multiple rules can be declared together on a node or relationship target.
  • block entries can mix primitive constraints already supported by the engine with runtime boolean validations.
  • added SHOW CONSTRAINT CONTRACTS to inspect stored contracts separately from the compiled primitive constraints they generate.

Example - node contract:

CREATE CONSTRAINT person_contract
FOR (n:Person)
REQUIRE {
  n.id IS UNIQUE
  n.name IS NOT NULL
  n.status IN ['active', 'inactive']
  (n.tenant, n.externalId) IS NODE KEY
  COUNT { (n)-[:PRIMARY_EMPLOYER]->(:Company) } <= 1
}

Example - relationship contract:

CREATE CONSTRAINT works_at_contract
FOR ()-[r:WORKS_AT]-()
REQUIRE {
  r.id IS UNIQUE
  r.startedAt IS NOT NULL
  startNode(r) <> endNode(r)
  startNode(r).tenant = endNode(r).tenant
  r.hoursPerWeek > 0
}

• Relationship cardinality and endpoint policy constraints (NornicDB extensions):
  • added directional relationship cardinality limits using REQUIRE MAX COUNT <n>.
  • added endpoint policy constraints so allowed start-label/end-label combinations can be enforced for a relationship type.

Example - directional cardinality:

CREATE CONSTRAINT max_employers
FOR ()-[r:WORKS_AT]->()
REQUIRE MAX COUNT 3

Changed

• Cypher parser compatibility and DDL coverage:

  • expanded ANTLR parser support for Neo4j 5-style REQUIRE forms, parenthesized property chains in key constraints, trailing OPTIONS, and WITH EMBEDDING syntax.
  • aligned parser-mode behavior so the same user-facing query shapes work consistently in both parser paths.

• Traversal planning and query execution:

  • added a generic relationship ORDER BY ... LIMIT planner that can seed traversal from the filtered end-node side using index top-k lookups.
  • added dedicated hot-path tracing for the new traversal planner and broadened routing/coverage for the translation-query family and adjacent query shapes.

• Release and dependency maintenance:

  • refreshed Go, UI, and workflow dependencies.
  • updated release/readme/docs content to reflect the current feature set and release flow.

Fixed

• Explicit transaction final-state semantics:

  • fixed explicit transaction handling so relationship constraints are validated against the final committed state rather than intermediate statement state.
  • fixed transaction read-your-own-write behavior for relationship reads and relationship property updates in explicit transactions.

• Large batch write amplification:

  • fixed UNWIND/bulk write paths that could fail with oversized transactions when a node was created and then updated again in the same transaction.
  • create-then-update mutations now collapse into a single pending create operation where possible.

• Relationship query ordering and stability:

  • fixed ORDER BY handling for relationship query results when ordering by aliased return expressions or node properties.
  • fixed MATCH-clause ORDER BY extraction so labels and relationship types such as :Order and :ORDERS no longer interfere with sorting.
  • restored deterministic ordering for Northwind-style relationship aggregation fast paths.

Tests

• Added and expanded regression coverage for:
  • block-style constraint contracts and SHOW CONSTRAINT CONTRACTS
  • relationship cardinality and endpoint policy constraints
  • explicit transaction relationship final-state validation
  • batch UNWIND create/update collapse behavior
  • exact translation query-family parsing, routing, e2e behavior, and traversal hot-path assertions
  • relationship aggregation ordering and ORDER BY clause parsing around Order/ORDERS tokens

Technical Details

• Range covered: v1.0.38..HEAD
• Commits in range: 19 (non-merge)
• Repository delta: 85 files changed, +7,997 / -2,854 lines
• Primary focus areas: schema contracts and relationship policy constraints, parser compatibility, explicit transaction correctness, bulk-write stability, and generic relationship traversal/query hot paths.

v1.0.38 - Phantom

EDIT: This was previously called "dimes." because I was just really releasing "my own stuff," I had started releasing versions with music I was currently listening to when I hit the button to release that version. However, this is starting to see real traction potentially, especially in corporate and academic settings, so going forward all songs will be more suitable for a corporate setting. Thanks for your support!

Phantom brings full Neo4j relationship constraint parity — uniqueness, existence, property type, and relationship key constraints now work on relationships using FOR ()-[r:TYPE]-() syntax, with automatic owned backing indexes, creation-time validation against existing data, and enforcement across every write path including transactions and bulk operations.

Phantom also introduces two NornicDB-exclusive constraint extensions: temporal no-overlap constraints on relationships that prevent overlapping time intervals grouped by a composite key, and domain/enum constraints that restrict any node or relationship property to a declared set of allowed values.

On the operational side, Phantom adds CMEK/HSM-backed at-rest encryption, fixes embedding property namespace pollution so the embedding property name is no longer reserved, stabilizes ORDER BY on returned node properties, aligns CREATE CONSTRAINT duplicate-name semantics with Neo4j by requiring IF NOT EXISTS for idempotent creation, and ships macOS installer service lifecycle fixes.

[v1.0.38] - 2026-04-02

Added

• Neo4j relationship constraint parity:
  • all constraint families now work on both nodes and relationships: uniqueness, existence (IS NOT NULL), property type (IS :: TYPE), and key constraints (IS RELATIONSHIP KEY).
  • relationship constraints use the FOR ()-[r:TYPE]-() target pattern, matching Neo4j 5.x syntax.
  • uniqueness and key constraints on relationships automatically create owned backing indexes that are dropped when the constraint is dropped.

Example — relationship uniqueness:

CREATE CONSTRAINT transfer_ref_unique IF NOT EXISTS
FOR ()-[r:TRANSFERRED]-()
REQUIRE r.reference_id IS UNIQUE

Example — relationship existence:

CREATE CONSTRAINT transferred_amount_exists IF NOT EXISTS
FOR ()-[r:TRANSFERRED]-()
REQUIRE r.amount IS NOT NULL

Example — relationship property type:

CREATE CONSTRAINT transferred_amount_type IF NOT EXISTS
FOR ()-[r:TRANSFERRED]-()
REQUIRE r.amount IS :: FLOAT

Example — relationship key (composite uniqueness + existence):

CREATE CONSTRAINT works_at_key IF NOT EXISTS
FOR ()-[r:WORKS_AT]-()
REQUIRE (r.employee_id, r.department) IS RELATIONSHIP KEY

• Temporal no-overlap constraints on relationships (NornicDB extension):
  • extends the existing node temporal no-overlap constraint to relationships.
  • supports a composite endpoint-pair form where the last two properties are always the time range and any preceding properties form the grouping key.

Example — 3-property form (single grouping key):

CREATE CONSTRAINT employment_temporal IF NOT EXISTS
FOR ()-[r:WORKS_AT]-()
REQUIRE (r.employee_id, r.valid_from, r.valid_to) IS TEMPORAL NO OVERLAP

Example — 4-property form (composite grouping key):

CREATE CONSTRAINT assignment_temporal IF NOT EXISTS
FOR ()-[r:ASSIGNED_TO]-()
REQUIRE (r.employee_id, r.project_id, r.valid_from, r.valid_to) IS TEMPORAL NO OVERLAP

• Domain/enum constraints (NornicDB extension):
  • new constraint type that restricts a property to a fixed set of allowed values.
  • works on both nodes and relationships.
  • domain constraints with different allowed-value sets on the same schema are treated as conflicting.

Example — node domain:

CREATE CONSTRAINT person_status_domain IF NOT EXISTS
FOR (n:Person)
REQUIRE n.status IN ['active', 'inactive', 'suspended']

Example — relationship domain:

CREATE CONSTRAINT works_at_role_domain IF NOT EXISTS
FOR ()-[r:WORKS_AT]-()
REQUIRE r.role IN ['engineer', 'manager', 'director']

• CMEK/HSM-backed at-rest encryption:

  • added customer-managed encryption key (CMEK) and HSM-backed at-rest encryption for the database engine.

• macOS installer service lifecycle:

  • added macOS packager workflow with automatic release upload.
  • separated macOS packager from Docker CD pipeline.

Changed

• IF NOT EXISTS semantics for CREATE CONSTRAINT:

  • exact-duplicate CREATE CONSTRAINT now returns an error unless IF NOT EXISTS is specified, matching Neo4j behavior. Previously, duplicates were silently idempotent.

• SHOW INDEXES and SHOW CONSTRAINTS relationship awareness:

  • SHOW INDEXES now reports entityType (NODE or RELATIONSHIP) and owningConstraint for backing indexes created by relationship uniqueness/key constraints.
  • SHOW CONSTRAINTS now reports entityType as RELATIONSHIP for relationship-scoped constraints.

• Branding and documentation:

  • abstracted Mimir-specific references to NornicDB-focused language across docs, source comments, test files, and configuration.
  • updated system-design documentation with WITH EMBEDDING support and correct embedding defaults (8192 chunk size, bge-m3 model).
  • updated user-facing documentation for all new constraint types with syntax examples.

Fixed

• Embedding property namespace pollution:

  • removed special-case routing that intercepted the embedding property name during SET/GET operations. The embedding property is now treated as a regular user property. Users can name their embedding fields anything and create vector indexes for them.

• ORDER BY on returned node properties:

  • stabilized ORDER BY when sorting on properties of returned nodes (e.g., ORDER BY t.createdAt when only t is returned).
  • fixed row-sorting to resolve var.prop against map-backed node values the same way expression evaluation does.

• Constraint enforcement namespace filtering:

  • fixed empty-namespace prefix filtering that caused unnamespaced edge IDs to be skipped during constraint enforcement scans.
  • fixed cross-namespace leak in transaction-level temporal constraint enforcement where pending edges from other namespaces could cause false violations.

• Owned backing index metadata:

  • fixed SHOW INDEXES to carry entityType and owningConstraint through relationship-scoped backing indexes instead of hardcoding NODE and nil.

• macOS installer service lifecycle:

  • fixed start/stop/restart controls to behave correctly for the macOS service.

• Graph API hardening:

  • hardened graph MVCC/diff handlers, aligned graph endpoint contracts, and improved validation and error mapping for the graph API.

Security

• Updated dependencies to resolve vulnerability scan findings.
• Updated example code to pass security scanning.

Tests

• Added and expanded regression coverage for:
  • all relationship constraint families (uniqueness, existence, property type, key) with DDL, creation-time validation, and write-path enforcement tests
  • relationship temporal no-overlap constraints (3-property and 4-property forms)
  • domain/enum constraints on nodes and relationships with conflict detection
  • IF NOT EXISTS idempotency semantics for all constraint types
  • embedding property namespace pollution removal
  • ORDER BY stability on returned node properties
  • CMEK/HSM encryption compliance
  • expanded unit test coverage across uncovered packages

Technical Details

• Range covered: v1.0.37..HEAD
• Commits in range: 30 (non-merge)
• Repository delta: 180 files changed, +15,077 / -2,619 lines
• Primary focus areas: Neo4j relationship constraint parity, NornicDB constraint extensions (temporal no-overlap, domain/enum), CMEK encryption, embedding property namespace cleanup, ORDER BY stability, and macOS packaging.