Ai

MemPalace v3.3.1 Improves Multilingual Entity Detection and Memory Graph Stability

MemPalace v3.3.1 Improves Multilingual Entity Detection and Memory Graph Stability

MemPalace v3.3.1 delivers a focused update centered on multilingual accuracy, safer local-first behavior, and stronger runtime stability. This release expands entity detection to five additional locales, improves name handling across non-Latin scripts, and tightens reliability in knowledge graph and MCP-related workflows, making the platform more dependable for AI memory and structured entity extraction use cases.

What Changed

One of the biggest additions in v3.3.1 is expanded multi-language entity detection. The release introduces full entity-detection patterns for Portuguese, Russian, Italian, Hindi, and Indonesian, broadening MemPalace's usefulness for global datasets and multilingual AI workflows.

The update also improves script-aware word boundaries, addressing name truncation issues in Devanagari, Arabic, Hebrew, Thai, Tamil, and Khmer scripts. This is a meaningful quality improvement for teams processing names, entities, and references across diverse writing systems where simple Latin-centric tokenization often fails.

Language handling is also more forgiving in day-to-day usage. MemPalace now supports case-insensitive BCP 47 language codes, so inputs such as PT-BR, zh-cn, and Pt-Br resolve correctly without manual normalization.

On the systems side, v3.3.1 strengthens knowledge graph thread safety by adding locking protections around close(), query_relationship, timeline, and stats. That should reduce race-condition risk in concurrent or automation-heavy environments.

Privacy and operational safety also get attention in this release. entity_registry.research() now defaults to local-only behavior, which means there are no outbound Wikipedia calls unless users explicitly opt in. The precompact hook has also been adjusted so failures or timeouts no longer block compaction entirely.

Additional hardening includes MCP stdout redirection to prevent library logging from corrupting the JSON-RPC channel, plus tighter file permission controls on sensitive palace data.

Why It Matters

For teams using AI memory systems, entity extraction, or structured knowledge workflows, v3.3.1 is less about flashy new surface features and more about making core operations safer and more robust. The multilingual additions directly improve extraction quality across more regions and scripts, which is increasingly important as AI applications expand beyond English-first environments.

The local-only default for research behavior is also notable. It reflects a stronger privacy posture and gives developers clearer control over outbound enrichment behavior, which matters in enterprise and regulated deployments.

At the same time, the thread-safety fixes and MCP logging protections target the kinds of bugs that can quietly undermine production reliability. For developers integrating MemPalace into larger AI pipelines, these low-level fixes may have an outsized impact on stability and trustworthiness.

Overall, MemPalace v3.3.1 looks like a practical, production-minded update. It improves multilingual entity detection, reduces operational edge cases, and strengthens the safety defaults that matter when memory infrastructure is handling sensitive or business-critical data.

Official Source: https://github.com/MemPalace/mempalace/releases/tag/v3.3.1