mnesis.models.summary

summary

Summary DAG and file reference data models.

SummaryNode

Bases: BaseModel

A materialized summary of a contiguous span of messages.

At MVP level (Phase 1) these are stored as assistant messages with is_summary=True in the ImmutableStore. The SummaryDAGStore adapts those messages into this model.

Two kinds exist:

• "leaf" — produced by summarising raw messages directly.
• "condensed" — produced by condensing one or more existing summary nodes; parent_node_ids records all consumed node IDs.

id instance-attribute

id: str

Same as the corresponding message ID in the ImmutableStore.

level class-attribute instance-attribute

level: int = 0

DAG depth: 0 = covers raw messages directly, 1+ = covers other nodes.

kind class-attribute instance-attribute

kind: Literal['leaf', 'condensed'] = 'leaf'

'leaf' = summarised from raw messages; 'condensed' = merged from summary nodes.

span_start_message_id instance-attribute

span_start_message_id: str

ID of the first message in the summarized span.

span_end_message_id instance-attribute

span_end_message_id: str

ID of the last message in the summarized span (inclusive).

content instance-attribute

content: str

The summary text that will be prepended to the active context.

parent_node_id class-attribute instance-attribute

parent_node_id: str | None = None

ID of the parent SummaryNode in a multi-level DAG (kept for backwards compat).

parent_node_ids class-attribute instance-attribute

parent_node_ids: list[str] = Field(default_factory=list)

All parent SummaryNode IDs consumed by this condensation (populated for 'condensed').

compaction_level class-attribute instance-attribute

compaction_level: int = 1

Escalation level: 1 = selective, 2 = aggressive, 3 = deterministic (fallback).

MessageSpan

Bases: BaseModel

A contiguous span of messages not yet covered by any SummaryNode.

FileReference

Bases: BaseModel

Content-addressed external file reference stored in the file_references table.

Used by LargeFileHandler to avoid inlining large files into the context window. Keyed by content hash so identical files are deduplicated automatically.

content_id instance-attribute

content_id: str

SHA-256 hex digest of the file content.

path instance-attribute

path: str

Original file path as seen by the application.

file_type instance-attribute

file_type: str

Detected MIME type or language identifier.

token_count instance-attribute

token_count: int

Estimated token count of the full file.

exploration_summary instance-attribute

exploration_summary: str

Compact structural description generated by LargeFileHandler.