Skip to content
Portable AI Memory

Schema Overview

PAM uses three JSON Schema (Draft 2020-12) files to define its data model:

SchemaFileStatus
Memory Storeportable-ai-memory.schema.jsonRequired
Conversationsportable-ai-memory-conversation.schema.jsonOptional
Embeddingsportable-ai-memory-embeddings.schema.jsonOptional

Architecture

Section titled “Architecture”
Required Memory Store portable-ai-memory.schema.json memories · owner · relations · conversations_index · integrity confidence · access · metadata · signature conversations_index embedding_ref Optional Conversations ...-conversation.schema.json messages · participants · provider · DAG Optional Embeddings ...-embeddings.schema.json vectors · model · dimensions

The Memory Store is the root document. It contains the memories array, owner information, metadata, relations, and an optional integrity block. Every valid PAM export must include a memory store file.

Conversations are companion files containing normalized chat history imported from providers. The memory store can reference conversations via the conversations_index field, linking memories to the conversations they originated from.

Embeddings are optional companion files containing vector representations of memory content. They are separated by design to keep the core memory store lightweight — systems that don’t need semantic search can ignore them entirely.

All three schemas comply with JSON Schema Draft 2020-12 and can be validated using any compliant validator (e.g., ajv-cli with --spec=draft2020).

Download

Section titled “Download”