v1.0.0 — Initial Release

Alaz v1.0.0

Long-term memory for AI coding agents.

Highlights

• 6-signal hybrid search — FTS + Dense Vector + ColBERT MaxSim + Graph Expansion + RAPTOR + Memory Decay, fused via Reciprocal Rank Fusion
• Autonomous learning pipeline — Extracts patterns, episodes, procedures, and core memories from session transcripts automatically
• Knowledge graph — Polymorphic graph with 14+ relation types, causal chains, cross-project promotion
• 22 MCP tools via StreamableHTTP for seamless Claude Code integration
• RAPTOR clustering — Hierarchical K-Means++ with LLM-generated summaries for conceptual search
• ColBERT token-level search — Jina-ColBERT-v2 for precise multi-vector retrieval
• Contradiction detection — Automatically detects and supersedes outdated knowledge
• Encrypted vault — AES-256-GCM secret storage with per-owner isolation
• Session checkpointing — Save and restore arbitrary session state
• HyDE — Hypothetical Document Embeddings for improved recall
• Circuit breaker — Graceful degradation when external services are unavailable

Tech Stack

• Rust (2024 edition) — Single binary, 9 crates, ~31K lines
• Axum — HTTP REST + MCP server
• SQLx — PostgreSQL with 18 migrations
• Qdrant — Dense (4096-dim) + ColBERT (128-dim) vectors
• Tokio — Async runtime, all search signals concurrent

Quick Start

git clone https://github.com/Nonanti/Alaz.git
cd Alaz
docker compose up -d
cp .env.example .env
cargo run -- migrate
cargo run -- serve

See README for full documentation.