v0.7.3

A token-cost release for agents using SLayer over MCP: search and model summaries are now compact by default.

Smaller default responses

search, models_summary, and inspect_model now return short description previews instead of full bodies. On agent workloads this cuts MCP token cost by roughly 25%. Pass compact=False (or slayer search --verbose) to get the verbose pre-0.7.3 shape.

Memory descriptions

You can now attach an optional short summary (up to 500 characters) to each memory you save. Pass --description to slayer memory save, or set the description field on MCP save_memory / POST /memories. Memory descriptions are indexed for both lexical and embedding search, and they're what surfaces in compact search hits.

Datasource descriptions are now searchable

DatasourceConfig.description is indexed for lexical and embedding search, and edit_datasource refreshes the embedding when you change the description (best-effort if the embedding provider is unreachable). You can also write Cypher pre-filters against Memory.description, Datasource.description, and Aggregation.description.

Bug fixes

slayer datasources create demo no longer crashes on Windows when jafgen emits non-Latin-1 glyphs.

Dialect-specific code has been consolidated under slayer/sql/dialects/. If you previously imported the SQLite UDF helper from slayer.sql.sqlite_udfs, update to from slayer.sql.dialects.sqlite import register_sqlite_udfs.