"""
agentmemory — a minimal, dependency-free agent memory reference implementation.

This package ships alongside the public reference site aiagentmemory.com.de as a
small, runnable demonstration of the core ideas behind "agent memory":

  1. An append-only episodic store (the raw record of what an agent did/saw).
  2. A semantic/retrieval layer that turns stored text into queryable memory.
  3. A provenance ledger so every returned memory can be traced to an original
     observation (no invented citations, no silent rewriting).
  4. A consolidation step that summarizes repeated facts into durable knowledge.

It is deliberately dependency-free (pure Python standard library) so the examples require no third-party installs. Python 3.14 was verified for
this release. The seeded fixture reproduces reported metric values; complete
JSON byte identity and compatibility with older runtimes are not established.

The bundled evaluation (``python -m agentmemory.eval``) uses a small, seeded,
self-contained dataset and prints a metrics report. Results are real outputs of
that command — they are *not* a claim about a proprietary benchmark.

Topics covered in the accompanying site:
    - /architecture/  -> the four-layer memory stack in this package
    - /quickstart/    -> install/run and 5-minute API walkthrough
    - /evaluation/    -> the reproducible retrieval evaluation + actual results
    - /examples/      -> save/retrieve/provenance/consolidation runnable examples
"""

from .mem import Entry, MemoryStore
from .retrieval import Retrieval
from .ledger import ProvenanceLedger
from .consolidate import Consolidator

__version__ = "0.1.0"

__all__ = [
    "Entry",
    "MemoryStore",
    "Retrieval",
    "ProvenanceLedger",
    "Consolidator",
    "__version__",
]