v0.6.0

Added

• reviewable memory synthesis proposal model in mnemara-core, including
  SynthesisRequest, SynthesisProposal, SynthesisReport, and
  MemoryStore::synthesize
• deterministic synthesis support in both file and sled stores for deriving
  draft Summary records from scoped source memories while preserving source
  lineage, review metadata, derived trust, and dry-run-first workflows
• HTTP /admin/synthesize and gRPC Synthesize surfaces, including protobuf
  request/reply messages, server mappings, request traces, and synthesis
  metrics
• JavaScript and Python SDK helpers for calling /admin/synthesize
• maintenance-run synthesis integration through run_synthesis, synthesis
  reports in maintenance responses, and scheduled dry-run synthesis via
  MNEMARA_BACKGROUND_MAINTENANCE_SYNTHESIS
• backend, transport, and SDK regression coverage for synthesis proposal
  generation, dry-run behavior, persisted proposal lineage, and maintenance
  orchestration
• synthesis proposal benchmark slices and checked-in artifacts covering
  dry-run, filtered dry-run, and apply paths for file and sled stores

Changed

• Repository docs, benchmark methodology/results, and crate READMEs now describe
  synthesis as reviewable summary proposal generation rather than autonomous
  memory rewriting.
• Marketing-site copy now mentions reviewable synthesis proposals alongside the
  shipped lifecycle, changefeed, SDK, and daemon capabilities.

Full Changelog: v0.5.0...v0.6.0

v0.5.0

What's Changed

Added

• judged recall evaluation harness types and helpers for expected, optional,
  disallowed, and explanation-note assertions
• additive TimeTravelRecallRequest / recall_as_of time-travel recall over
  backend-maintained record versions in the file and sled stores, exposed over
  HTTP /memory/recall-as-of, gRPC RecallAsOf, and the JavaScript/Python SDKs
• append-only changefeed request/report/event types plus file, sled, HTTP, and
  JavaScript/Python SDK access through /admin/changefeed

Changed

• Roadmap now promotes evaluation harness, changefeed/watch API, and
  time-travel recall into the active implementation plan and records the next
  candidate roadmap items.

Full Changelog: v0.4.0...v0.5.0

v0.4.0

Added

• platform/runtime expansion: RunMaintenance core, HTTP, and gRPC admin operation that orchestrates integrity checks, idempotency-key repair, and tenant-scoped compaction
• daemon background maintenance scheduler controlled by MNEMARA_BACKGROUND_MAINTENANCE_* environment variables
• daemon-to-daemon snapshot shipping through portable export/import semantics, exposed as POST /admin/replication/ship and gRPC ShipSnapshot
• mnemara-ffi crate with a JSON-based C ABI over the sled-backed store for FFI-capable runtimes
• dependency-free Python HTTP SDK covering memory, recall, graph inspection, integrity, repair, maintenance, traces, runtime status, portable export/import, and snapshot shipping
• JavaScript SDK helpers and type declarations for maintenance runs and snapshot shipping
• release checklist sdk-package and sdk-publish phases for JavaScript and Python SDK validation plus credentialed npm/PyPI publication
• end-to-end tests for HTTP snapshot shipping, gRPC maintenance orchestration, FFI upsert/recall/maintenance, Python SDK request handling, and background maintenance environment parsing

Changed

• Platform and Runtime Expansion roadmap entries are now marked shipped, with deployment, user-guide, SDK, FFI and server README updated to match the implemented behavior
• JavaScript SDK package metadata now aligns with the workspace 0.3.0 release line and includes public publish configuration

Full Changelog: v0.3.0...v0.4.0

v0.3.0

Added

• additive conflict-review metadata on memory records, including drift score, conflict state, related record ids, operator resolution, resolver, resolution timestamp, and notes
• recall filters for relative before_record_id / after_record_id temporal anchors, boundary labels, recurrence keys, conflict states, resolution kinds, and unresolved conflict review queues
• conflict-aware recall explanation channels and regression coverage for relative episodic retrieval, boundary filtering, and conflict-review recall
• JavaScript SDK constants and type declarations for conflict-review and resolution enums
• embeddable animated memory-engine visualization for README and marketing-site surfaces

Changed

• HTTP memory and recall routes now enforce the same semantic request limits as gRPC for batch breadth, recall breadth, labels, content, and summaries
• batch upsert now pre-validates all records before applying writes, avoiding partial mutation when a later item is invalid

Fixed

• replacing a record with a different idempotency key now removes the old scoped idempotency mapping instead of leaving a stale retry key behind
• memory record validation now rejects out-of-range importance_score values

Full Changelog: v0.2.1...v0.3.0

v0.2.1

What's Changed

• Expose shared semantic embedder injection for embedded stores with benchmarked custom-provider seam by @Copilot in #2

New Contributors

• @Copilot made their first contribution in #2

Full Changelog: v0.2.0...v0.2.1

v0.2.0

Added

• additive episodic record fields for continuity state, salience, causal links, and optional affective annotations
• retrieval-planner profiles with planner-stage and candidate-source traces for explainable continuity-aware recall
• lifecycle-aware historical-state and lineage-link model fields across the core, protocol, daemon, and backend layers
• compaction and maintenance counters for superseded records, historical records, and lineage links
• workload-aware retrieval policy profiles for general, support, research, assistant, and autonomous-agent recall tuning
• public shared-embedder constructors for custom semantic providers in the scorer and planner path

Changed

• recall explanations now expose planning profile, episodic and salience score channels, and richer planner trace detail
• recall explanations and engine tuning exports now disclose the effective retrieval policy profile alongside scorer and planner metadata
• custom semantic provider notes now flow through explanation policy notes on the same path as built-in embedding providers
• compaction and archival flows now preserve current versus historical versus superseded visibility instead of treating all archived data as the same state
• repository docs, deployment guidance, benchmark docs, and website rollout copy were aligned with shipped episodic, planner, and lifecycle behavior
• benchmark report v1 now publishes the expanded 16-case corpus, lifecycle-sensitive scenario tables, and the continuity-aware planner profile slice

Fixed

• archived replay and roundtrip coverage now explicitly exercises historical recall visibility after lifecycle-aware maintenance transitions
• daemon startup now honors the documented MNEMARA_RECALL_PLANNING_PROFILE and MNEMARA_GRAPH_EXPANSION_MAX_HOPS fallback controls, with regression coverage for safer rollout posture
• policy-tuned provenance ranking now applies the configured workload profile consistently across file, sled, daemon, and transport explanation surfaces

Full Changelog: v0.1.0...v0.2.0

v0.1.0

Added

• facade crate mnemara with feature-gated re-exports for the file, sled, protocol, and server surfaces
• core domain model and async store traits in mnemara-core, including scoped memory records, recall queries, explanations, planning traces, and maintenance report types
• embedded file-backed and sled-backed storage implementations with batch upsert, snapshot, delete, compaction, integrity check, repair, export, and import support
• protobuf and gRPC API surface in mnemara-protocol for Rust client and server integration
• standalone daemon in mnemara-server with tonic gRPC service plus HTTP/JSON admin endpoints
• reference JavaScript HTTP SDK for non-Rust consumers
• explainable recall with per-hit score breakdowns, selected channels, planning traces, and correlation IDs
• configurable recall scorer kinds and scoring profiles for balanced, lexical-first, and importance-first retrieval behavior
• semantic embedding seam with a deterministic local provider for offline usage and integration testing
• ranking evaluation assets and benchmark artifacts documenting quality and backend parity
• retry-safe idempotent writes and batch upserts
• tombstone and hard-delete flows for memory records
• duplicate-aware compaction with rollup summaries and optional cold-tier archival for stale low-importance records
• retention controls for TTL expiry, archival windows, and namespace caps
• backend-neutral export and import packages with validate, merge, replace, and dry-run flows across file and sled stores
• daemon deployment profiles for TCP, Unix domain sockets, TLS, and mutual TLS
• request-size and batch-size guardrails plus bounded admission control with tenant-aware fairness
• metrics export, request traces, runtime status visibility, and trace filtering APIs
• HTTP admin surfaces for health, readiness, snapshot, stats, integrity, repair, compact, delete, trace lookup, runtime status, export, and import operations
• architecture, deployment, user guide, benchmark methodology, benchmark results, ranking defaults ADR, and security policy documentation
• release checklist script and crates.io README audit for publication readiness
• installation and publishing guidance for source checkouts, published crates, and the daemon binary

Full Changelog: https://github.com/deliberium/mnemara/commits/v0.1.0