Roadmap

Completed

v0.8.0 - RAG Integration (Current)

• [x] RAG Bridge - Accept external RAG chunks, enhance with memory signals
• [x] Memory Enhancer - Enrich retrieval with heuristics, anti-patterns, domain knowledge
• [x] Retrieval Feedback Loop - Learn which retrieval strategies work
• [x] IR Metrics - MRR, NDCG, Recall, Precision, MAP (pure Python)
• [x] Hybrid Search - Vector + keyword (BM25/TF-IDF) with RRF fusion
• [x] Cross-Encoder Reranking - Pluggable reranker with optional rerankers library
• [x] Text Search Providers - BM25S + pure-Python TF-IDF fallback
• [x] Optional dependency group: pip install alma-memory[rag]

v0.7.0 - Intelligence Layer

• [x] Memory Decay - Time-based confidence decay for aging memories
• [x] Memory Compression - LLM + rule-based memory summarization
• [x] Retrieval Verification - Two-stage verified retrieval pipeline
• [x] Retrieval Modes - 7 cognitive task modes (standard, progressive, verified, budget, etc.)
• [x] Trust-Integrated Scoring - Agent trust weighting for memory retrieval
• [x] Token Budget Management - Context window management
• [x] Progressive Disclosure - Summary, highlights, full detail levels
• [x] 6 New MCP Tools - Memory Wall tools for Claude Code
• [x] Archive System - Soft-delete with recovery

v0.6.0 - Workflow Context Layer

• [x] Checkpoint & Resume - Save and restore workflow state
• [x] State Reducers - Merge outputs from parallel agents
• [x] Artifact Linking - Track code, tests, and documents per workflow
• [x] Scoped Retrieval - Filter memories by workflow context
• [x] Session Persistence - Handoffs survive process restarts
• [x] 8 New MCP Tools - Full workflow support in Claude Code
• [x] TypeScript SDK v0.6.0 - Full workflow API parity

v0.5.0 - Vector Database Backends

• [x] Qdrant Backend - Managed vector database support
• [x] Pinecone Backend - Serverless vector database
• [x] Chroma Backend - Lightweight local option
• [x] Graph Database Abstraction - Neo4j, Memgraph, Kuzu backends
• [x] Testing Module - MockStorage, MockEmbedder, factories
• [x] Memory Consolidation - LLM-powered deduplication
• [x] Event System - Webhooks and in-process callbacks
• [x] TypeScript SDK - Initial release
• [x] Multi-Agent Memory Sharing - inherit_from, share_with

v0.4.0 and Earlier

• [x] Core memory types (Heuristic, Outcome, Preference, Knowledge, Anti-pattern)
• [x] Scoped learning (can_learn, cannot_learn)
• [x] SQLite + FAISS storage
• [x] PostgreSQL + pgvector
• [x] Azure Cosmos DB
• [x] MCP Server integration
• [x] Progress tracking
• [x] Session handoff
• [x] Domain memory factory
• [x] Confidence engine
• [x] Cache layer
• [x] Forgetting mechanism

In Progress

v0.9.0 - Visualization & Temporal (Planned)

• [ ] Visual memory explorer dashboard
• [ ] Memory graph visualization
• [ ] Analytics and insights
• [ ] Temporal reasoning (time-aware retrieval)
• [ ] Proactive memory suggestions

Future

• [ ] Additional graph backends (Neptune, TigerGraph)
• [ ] Ollama embedding support
• [ ] Groq LLM integration
• [ ] Memory import/export
• [ ] Team collaboration features

Contributing

Want to help? Check out our Contributing Guide and Good First Issues.