December 11: Support for GPT-5.2, Attio CRM, mirror sync mode, filter modes

New Features

• 💡 Graphlit now supports OpenAI GPT-5.2 with the model enum GPT52_400K.

• 💡 We have completed Attio CRM integration as a data source, allowing you to ingest companies, contacts, and custom CRM objects directly into your knowledge graph. Custom object mappings let you model your specific CRM structure (LPs, Investments, External Funders, etc.).

• We have added MIRROR sync mode for Azure blob and S3 feeds, enabling bidirectional synchronization where deletions in the source are reflected in Graphlit.

• We have added filter modes (any, all, only) for collections and other filter objects, providing more precise control over how multiple filter values are combined.

• We have added an entity relationship lookup query for efficient traversal of knowledge graph relationships without scanning the full graph.

• MCP server support is now available directly from Graphlit projects, similar to Zine. Developers can expose their project content via remote MCP with API key authentication.

• We have added a relevance threshold property to filter objects, allowing developers to override the default vector similarity threshold.

Bugs Fixed

• GPLA-4758: Fixed precondition failures when updating content concurrently

• GPLA-4764: Person entities now correctly generate vector embeddings

• GPLA-4763: Fixed specification upsert matching similar names incorrectly (e.g., "GPT-5" vs "GPT-5.1")

• GPLA-4724: Fixed slow CosmosDB queries with multi-partition fallback retries

• GPLA-4682: Credits quota exceeded now returns proper PaymentRequired status instead of BadRequest