ALMA-memory v0.8.0 — RAG Integration Layer

RAG Integration Layer

ALMA v0.8.0 adds a complete RAG integration layer that enhances any RAG framework with memory signals.

New Features

• RAG Bridge — Accept chunks from any RAG framework (LangChain, LlamaIndex, custom) and enhance with ALMA memory signals
• Hybrid Search — Vector + keyword search with Reciprocal Rank Fusion (RRF)
• Retrieval Feedback Loop — Track and auto-tune retrieval weights based on outcomes
• IR Metrics Engine — MRR, NDCG, Recall, Precision, MAP — pure Python, deterministic
• Cross-Encoder Reranking — Pluggable reranking pipeline
• Memory Consolidation — LLM-powered deduplication across memory types

Tech Debt Remediation (from v0.7.1)

• Split MCP tools god file (~3,000 lines) into 5 focused modules
• Added 155 tests for retrieval modules, found and fixed 3 latent bugs
• Embedding performance boost (2.6x faster via batched processing + LRU cache)
• Storage backend factory pattern for cleaner instantiation
• 15 cross-module integration tests

Documentation & Quality (v0.8.0+)

• Restructured .claude/ with AIOS patterns: 4 agents, 3 skills, 5 AIDR records
• 4 dark-theme architecture diagrams (learning cycle, memory types, multi-agent sharing, architecture)
• Replaced all internal agent persona names with descriptive roles across README + docs
• Landing page (alma-memory.pages.dev) updated with diagrams and v0.8.0 content

Install

pip install alma-memory                # Core
pip install alma-memory[rag]           # + RAG integration (hybrid search, reranking)
pip install alma-memory[all]           # Everything
npm install @rbkunnela/alma-memory     # TypeScript SDK

Stats

• 1,682 tests passing (118 new RAG tests)
• 115 Python files, 22 MCP tools, 7 storage backends, 4 graph backends
• 5 documented architectural decision records (AIDR)

Full changelog: CHANGELOG.md