v1.4.0

What's New

• Confidence scoring — Every search returns a confidence level (HIGH, MODERATE, LOW, NONE) based on result relevance
• Uncertainty modes — Three strategies for uncertain retrieval: strict (refuse), helpful (hedge), creative (offer to guess)
• mem.pin() — Store persistent memories with elevated importance that resist decay
• Frustration detection — Detects "I told you this!" patterns and provides recovery guidance
• Retrieval modes — fast, balanced, deep presets for accuracy-cost tradeoffs
• Query-adaptive scoring — Scoring weights adapt to query type (factual, temporal, multi-hop)
• Two-pass re-ranking — Better precision for factual queries
• MCP tools — widemem_pin and widemem_export added to MCP server
• Ollama embeddings — Fully local operation with Ollama as embedding provider

See CHANGELOG.md for full details.