Skip to content

RELEASE_NOTES.md

Latest commit

 

History

History
118 lines (91 loc) · 5.62 KB

RELEASE_NOTES.md

File metadata and controls

118 lines (91 loc) · 5.62 KB
Raw

Neuroca Release Notes

1.0.0 – General Availability

Release date: 2025-09-22

Highlights

  • Async-first memory orchestration – The MemoryManager and MemoryRetrieval flows were rebuilt to normalise tier selection, surface structured MemoryRetrievalResult payloads, and preserve metadata filters across working, episodic, and semantic memories.
  • Production vector search – A dedicated QdrantVectorBackend now powers similarity queries with deterministic UUID handling, batched CRUD operations, and regression coverage for metadata-aware searches.
  • Knowledge graph relationships – Long-term relationship management stores metadata bidirectionally, exposing create/update/delete helpers through the manager interface and validating behaviour against the in-memory and Neo4j backends.
  • Operational tooling – The asynchronous benchmark harness and end-to-end validation suite exercise the full memory stack through the modern CLI bootstrapper, providing release-ready performance signals.
  • Documentation refresh – README quick starts, backend guides, and API references were aligned with the async storage factory and new retrieval results so integrations reflect the shipped interfaces.

Breaking changes

  • The legacy MemoryRetrieval stub now returns MemoryRetrievalResult instances and enforces tier validation. Custom callers should update any tuple-based unpacking logic accordingly.
  • Storage backends integrate through StorageBackendFactory.create_storage() and the new BackendType.QDRANT option. Plugins that relied on create_backend() must migrate to the asynchronous factory APIs.
  • Memory model compatibility shims moved into dedicated modules within neuroca.memory.models. Direct imports from neuroca.memory.memory_items should be updated to the re-exported package paths.

Upgrade notes

  • Install optional extras (pip install neuroca[vector,test]) to pull in the Qdrant client library when enabling the production vector backend.
  • Refresh tier configuration files to register the desired vector and knowledge graph backends. The docs include sample configuration blocks for local and managed deployments.
  • Re-run the provided benchmarks or smoke tests after upgrading to validate the configured storage backends and confirm the async bootstrapper wiring.

1.0.0-rc1 – Release Candidate

Release date: 2025-09-19 (candidate)

Highlights

  • Memory manager stabilization: clean demo run with one search hit, no warnings.
  • Audit/events readiness: event bus compatibility (BaseEvent), tags metadata normalized; tests green.
  • Production configuration: added config/production.yaml; Docker defaults to prod (ENV/NCA_ENV).
  • Observability: Prometheus metrics publisher integrated; disabled by default in demo.
  • Security: Codacy CLI Trivy shows zero vulnerabilities for poetry.lock and requirements.txt after tightening constraints (httpx, aiohttp, transformers, requests, protobuf, starlette, torch, pydantic, scikit-learn, urllib3). LangChain moved to optional extra.

Breaking changes (planned for 1.0.0)

  • LangChain integration is optional by default; install with extras: pip install .[integrations].
  • Torch minimum version raised on supported Python versions.

Upgrade notes

  • Review any custom constraints; align to new minimums.
  • If using LangChain adapters, enable extras and validate workflows.

RC validation plan

  • Soak test with a coding agent for several days under realistic load (chat sessions, memory churn, consolidation/decay active) and monitor metrics/events.
  • Exercise CLI backup/restore in both SQLite and Postgres modes and validate recovery.
  • Run full integration and end-to-end suites; fix any regressions before cutting 1.0.0.

0.1.0b1 – Beta Preview

Release date: 2025-09-16

The 0.1.0b1 beta refresh delivers the first cohesive release of the unified Neuroca memory system. Highlights include:

  • ✅ Unified memory manager – A single async-first MemoryManager powers all tiers while preserving the legacy compatibility layer for synchronous integrations.
  • ✅ Vector search integration – Tier construction now provisions vector backends through StorageBackendFactory, enabling out-of-the-box similarity queries and long-term knowledge consolidation.
  • ✅ Async cognitive control – The decision maker, planner, and metacognitive monitor operate against the async manager, sharing utilities for deterministic option scoring and plan generation.
  • ✅ Expanded regression coverage – New unit and integration suites verify vector-backed search, tier maintenance, API routes, and compatibility shims across the package surface.
  • ✅ Developer experience upgrades – Smoke-tested demo scripts, restored async test infrastructure, and vendored pytest_asyncio support ensure the full test suite executes reliably from source checkouts.

Installation Notes

  • Install production dependencies with pip install neuroca.
  • For development and testing, install optional extras:
    • pip install neuroca[dev,test]
    • or via Poetry: poetry install --with dev,test

Upgrade Guidance

  • Regenerate configuration files if they reference the deprecated neuroca.core.memory module; the tiered manager is now exported from neuroca.memory.manager.
  • Update cognitive-control extensions to use the async helper utilities in neuroca.core.cognitive_control._async_utils.
  • Refresh local caches for vector indexes before running the maintenance workflow tests (tests/integration/memory/test_maintenance_workflow.py).

For historical releases and future updates, see the documentation portal at https://docs.neuroca.dev/.