Chat history arrays are a log, not memory. This walks through the LLM memory layer I build instead: DSPy signatures for extracting atomic facts, QDrant for per-user vector storage, and a maintenance agent that decides when an old fact gets updated or deleted.
WordPress is testing Gutenberg Content Guidelines, a machine-readable place to keep brand voice, accessibility and image rules so plugins and AI tools can read them instead of guessing from a PDF nobody opens.
The WP AI Client merge proposal would put a provider-agnostic AI API in WordPress 7.0: one PHP interface, a fluent prompt builder, WP_Error returns, and no outbound calls until a plugin configures a provider.
The Core AI team put out a Call for Testing for the AI Experiments plugin. Here is what is in it: MCP server support, markdown feeds, extra providers and AI request logging, plus the $400 API bill that convinced me logging is the part you cannot skip.
A million-token context window does not mean a model can reason across a million tokens. Recursive Language Models hand the context to a sandboxed Python REPL and let the model write code to partition and query it, which is how DSPy now implements the strategy.
WordPress 7.0 may merge the WP AI Client into Core, giving plugins one provider-agnostic API for calling models instead of each shipping its own SDK. Notes on the Connectors keychain, the package size argument, and what it changes for developers.
WordPress 6.9 shipped the Abilities API, and the MCP Adapter is what lets AI agents find those abilities and call them. Ahmad Wael covers flagging an ability for MCP, the two transport methods, custom servers, and keeping permission callbacks tight.
Basic GNNs treat every SKU connection the same. Heterogeneous Graph Transformers weight each edge type separately, which cut WAPE from 0.86 to 0.58 on an FMCG dataset. The post also shows the typed-edge table I use instead of flat post meta.
The January 2026 Towards Data Science pieces I actually used: Ryan Pegoud cutting LLM memory 84% with Triton fused kernels, ACE context engineering, and Hugo Lu on the ceilings at Databricks and Snowflake.
RoPE rotates query and key vectors instead of adding an index, so the distance between two tokens stays the same wherever the sequence sits. Covers the rotation intuition, why dimension pairs rotate at different speeds, and a PyTorch implementation for a custom transformer.