Crate anamnesis_adapter_memos 

Anamnesis adapter for MemOS (MemTensor/MemOS, Apache-2.0).

MemOS organizes durable memory as “MemCubes” — each cube is a directory that, when dumped, contains a flat layout:

<cube_dir>/
├── config.json                  # cube config
├── textual_memory.json          # textual memory items (we read this)
├── activation_memory.pickle     # KV cache, binary — skipped
└── parametric_memory.adapter    # LoRA weights, binary — skipped

The adapter walks a MemOS root (default ~/.memos/), finds every textual_memory.json, and emits one AnamnesisRecord per textual item.

memory_type → Anamnesis classification:

MemOS memory_type	Kind	Scope
WorkingMemory	Reference	Ephemeral
LongTermMemory	Fact	User
UserMemory	Preference	User
PreferenceMemory	Preference	User
OuterMemory	Reference	User
ToolSchemaMemory	Skill	Project
SkillMemory	Skill	Project
ToolTrajectoryMemory	Episode	Project
RawFileMemory	Reference	Project
(any other / missing)	Reference	User

Flat-backend free-form type heuristics (fact, event, opinion, procedure) are also honored as a secondary mapping.

Tombstones (status ∈ {archived, deleted, resolving}) are skipped.

Per §-1.2.2 the adapter is read-only.

Re-exports

pub use detector::MemosDetector;

pub use scanner::MemosScan;

pub use scanner::MemosTextItem;

Modules

detector

Detect a MemOS install.

normalizer

Normalize MemOS raw records into AnamnesisRecords.

scanner

Filesystem scanner for MemOS (MemTensor/MemOS, Apache-2.0).

Structs

MemosAdapter

The adapter.

MemosConfig

Adapter configuration.

Constants

ADAPTER_ID

Stable adapter identifier.

Functions

memos_adapter

Convenience constructor.