Releases: SpillwaveSolutions/agent-memory
Releases · SpillwaveSolutions/agent-memory
Release 2.4.0
What's Changed
- feat: v2.4 Phase 30 — Claude Code CLI Harness by @RichardHightower in #18
- feat: v2.4 Phase 31 — Gemini CLI Tests by @RichardHightower in #19
- feat: v2.4 Phase 32 — OpenCode CLI Tests by @RichardHightower in #20
- feat: v2.4.0 Headless CLI Testing milestone by @RichardHightower in #21
Full Changelog: v2.3.0...v2.4.0
v2.3.0 — Install & Setup Experience
Summary
- Install & setup documentation: quickstart, full guide, and agent setup docs for all 4 adapters
- Four wizard-style setup skills (install/configure/verify/troubleshoot) with confirmation flows
- Performance benchmark harness (
perf_bench) with p50/p90/p99 latency baselines - Benchmark docs with usage, outputs, and baseline update workflow
Phases
- Phase 28: Install & Configuration Skills + User Guides
- Phase 29: Performance Benchmarks
References
v2.2.0 — Production Hardening
Summary
- Production hardening milestone: wired remaining gRPC stubs and fixed ListAgents session_count
- Added agent attribution on TeleportResult/VectorTeleportMatch for cross-agent visibility
- Added 29 E2E tests across 7 files (core pipeline + advanced scenarios)
- Dedicated E2E CI job with per-test reporting; required for PR merge
Phases
- Phase 24: Tech debt cleanup (RPC wiring, session_count fix, agent fields)
- Phase 25: 10 core E2E tests (pipeline, BM25, vector, topic graph, grip provenance)
- Phase 26: 19 advanced E2E tests (multi-agent, degradation, error paths)
- Phase 27: Dedicated E2E CI job in GitHub Actions
References
v2.0.0 - Cognitive Layers
Agent Memory v2.0.0 - Cognitive Layers
This major release implements the complete v2.0 cognitive architecture with 5 new search and retrieval layers. The system now provides multi-modal memory access through hierarchical navigation, keyword search, semantic similarity, and topic-based discovery.
Highlights
- 5 New Crates adding 11,208 lines of Rust code
- 578 Tests with comprehensive coverage
- 6-Layer Cognitive Architecture fully operational
- CI/CD Pipelines for automated testing and releases
New Features
Phase 10.5: Agentic TOC Search
SearchNodeRPC: Search within a specific TOC nodeSearchChildrenRPC: Recursive search through child nodesterm_overlap_score(): Index-free term matching algorithm- CLI:
query search --node <id> --query <terms>
Phase 11: BM25 Teleport (Tantivy)
- memory-search crate: Full-text search with Tantivy
TeleportSearchRPC: Keyword search with BM25 rankingGetTeleportStatusRPC: Index availability check- Background index commit job
- CLI:
query teleport-search --query <keywords>
Phase 12: Vector Teleport (HNSW)
- memory-embeddings crate: Candle ML with all-MiniLM-L6-v2 (384-dim)
- memory-vector crate: usearch HNSW vector index
VectorTeleportRPC: Semantic similarity searchHybridSearchRPC: RRF fusion (k=60) of BM25 + vectorGetVectorIndexStatusRPC: Index statistics- Model caching in
~/.cache/memory-embeddings - CLI:
query vector-teleport,query hybrid-search
Phase 13: Outbox Index Ingestion
- memory-indexing crate: Unified outbox consumer pipeline
- Checkpoint-based crash recovery
- Incremental index updates (add/update documents)
- Admin command:
admin rebuild-indexes - Scheduled background indexing job
Phase 14: Topic Graph Memory
- memory-topics crate: HDBSCAN density-based clustering
- LLM-generated labels with TF-IDF keyword fallback
- Exponential time-decay importance scoring (30-day half-life)
- Topic relationships: SIMILAR, PARENT, CHILD, SEQUENTIAL
GetTopicGraphStatus,GetTopicsByQueryRPCsGetRelatedTopics,GetTopTopicsRPCs- Lifecycle management: pruning dormant topics, resurrection
- CLI:
query topics,admin topic-prune
New gRPC RPCs
| RPC | Description |
|---|---|
SearchNode |
Search within a TOC node |
SearchChildren |
Recursive child search |
TeleportSearch |
BM25 keyword search |
VectorTeleport |
Semantic vector search |
HybridSearch |
Combined BM25 + vector |
GetVectorIndexStatus |
Vector index stats |
GetTopicGraphStatus |
Topic graph stats |
GetTopicsByQuery |
Query topics |
GetRelatedTopics |
Related topic discovery |
GetTopTopics |
Top topics by importance |
New Crates
| Crate | LOC | Description |
|---|---|---|
| memory-search | 1,470 | Tantivy BM25 full-text search |
| memory-embeddings | 571 | Candle ML embeddings |
| memory-vector | 1,381 | usearch HNSW index |
| memory-indexing | 2,584 | Outbox consumer pipeline |
| memory-topics | 5,202 | Topic clustering and relationships |
Infrastructure
- GitHub Actions CI: Format, clippy, test, build, doc checks
- Release Workflow: Multi-platform builds (Linux x86_64/ARM64, macOS Intel/Apple Silicon, Windows x86_64)
Breaking Changes
None - this release maintains backward compatibility with v1.0.0 APIs.
Upgrade Notes
The new crates are optional and integrate seamlessly with existing deployments. To enable the new features:
- Update to v2.0.0 binaries
- Indexes will be built automatically on first run
- Use new CLI commands or gRPC RPCs to access features
Full Changelog
v1.0.0 MVP - Agent Memory System
Agent Memory v1.0.0
Complete conversational memory system with TOC-based agentic navigation.
Key Features
- Storage: RocksDB with 6 column families, time-prefixed keys, crash recovery
- TOC Hierarchy: Year → Month → Week → Day → Segment with rollup summarization
- Grips: Provenance system linking summaries to source evidence
- gRPC Service: IngestEvent, GetTocRoot, GetNode, BrowseToc, GetEvents, ExpandGrip
- Claude Code Plugin: 3 commands + memory-navigator agent (99/100 skill grade)
- CCH Integration: memory-ingest binary with fail-open behavior
Stats
| Metric | Value |
|---|---|
| Phases | 8 |
| Plans | 20 |
| Files | 91 |
| Lines | 9,135 |
| Requirements | 50 |
What's Next (v2.0)
- Teleport indexes (BM25/vector search)
- Additional hook adapters (OpenCode, Gemini CLI)
- Production hardening