kestrel-feature-story-archive 0.3.0

Schema-rich agent life-story archive for Kestrel Sovereign

kestrel-feature-story-archive

Schema-rich agent life-story archive for Kestrel Sovereign.

An agent maintains a queryable, semantic archive of its life-story: interactions, observations, and reflections, organized as a timeline with vector search and contradiction detection. This is a reusable framework capability — a sibling package alongside kestrel-feature-healthcare / -visual / -reflection / -observability — not a host-product feature. Any Kestrel host can consume it.

Status

Bootstrap skeleton (FEAT-1). Business logic lands in FEAT-2..5:

• FEAT-2 — entity models (Story, StoryChunk, StoryEvent) + Alembic migrations
• FEAT-3 — TimelineBuilder (session → Story/StoryEvent/StoryChunk extractor)
• FEAT-4 — VectorSearch integration for semantic story retrieval
• FEAT-5 — sharing, replay, export, contradiction detection services

This package migrates the 8,240-LOC Story Archive stack from Frinz into a reusable feature package so other consumers (RemoteCares/RPM Azure pilot) can leverage it. Records are scoped to agent_did (the canonical owner identity per the Agent Identity Contract).

Installation

uv pip install kestrel-feature-story-archive

The package registers StoryFeature through the kestrel_sovereign.features entry-point group and registers its entity models via the kestrel_entities.models entry-point for Alembic autogen.

Development

uv sync --extra test
uv run --extra test pytest

Architecture

Records are agent_did-scoped and persisted via kestrel-feature-entities ORM layer. The feature consumes the DatabaseBackend from the host and leverages the Timeline and VectorSearch protocols (kestrel-sovereign-sdk v0.16.0+).

No product-specific terminology anywhere in this package — vocabulary is agent / agent_did / owner throughout. This is a framework capability.