summarization_model cognify knob as
CognifyConfig.summary_schema, consumed by the standalone cognify pipeline. Its
graph-extraction counterpart is the
custom graph schema.
What it does
Replaces the defaultSummarizedContent shape requested from the LLM during the
summarization stage with your own JSON Schema. Mirrors Python’s
CognifyConfig.summarization_model.
Requirement
The schema must contain a stringsummary property — the pipeline reads
summary to build each TextSummary. This is validated up front by
validate_summary_schema, so a bad schema
fails at config time rather than mid-pipeline.
Example (programmatic)
SummaryExtractor::new_with_schema(llm, config.summary_schema)
(crates/cognify/src/tasks.rs), and the
extractor requests the custom schema when it is Some
(crates/cognify/src/summarization/extractor.rs).
Via top-level config
cognee exposes a runtime setter mirroring Python’s
cognee.config.set_summarization_model(...):
Settings::cognify_config applies it, so every cognify run started from these
settings requests this shape. It can be set only programmatically or through a
binding’s set: it has no env var and is not saved, so cognee-cli cannot set it.
Pointers
CognifyConfig—summary_schemaandvalidate_summary_schema.crates/cognify/src/summarization/extractor.rs— summary schema consumption.