API Reference

Organization Ontology

Read and extend the memory vocabulary: the content types a memory can be classified as, and the relation types an edge between memories can carry.

Endpoints

GET/memory/ontology

200 OK

PUT/memory/ontology

200 OK

Read the current vocabulary

import os
from memorysync import MemorySyncClient

client = MemorySyncClient(
    api_key=os.environ["MEMORYSYNC_API_KEY"],
    base_url="https://api.memorysync.io",
    project_id=os.environ["MEMORYSYNC_PROJECT_ID"],
    end_user_id="usr_7f3a9c2e",
)

ontology = client.get_ontology()
print(ontology.content_types)
print(ontology.custom_relation_types)

200-response.json

{"content_types":["preference","fact","behavior","intent","temporal","relationship","skill","event","other"],"relation_types":["similar","derived_from","continuation","contradiction","summary_of","detail_of","caused_by","references","supports","extends"],"builtin_content_types":["preference","fact","behavior","intent","temporal","relationship","skill","event","other"],"builtin_relation_types":["similar","derived_from","continuation","contradiction","summary_of","detail_of","caused_by","references","supports","extends"],"custom_content_types":[],"custom_relation_types":[],"max_custom_types":32}

Add your own types

import os
from memorysync import MemorySyncClient

client = MemorySyncClient(
    api_key=os.environ["MEMORYSYNC_API_KEY"],
    base_url="https://api.memorysync.io",
    project_id=os.environ["MEMORYSYNC_PROJECT_ID"],
    end_user_id="usr_7f3a9c2e",
)

ontology = client.update_ontology(
    content_types=["symptom", "dosage"],
    relation_types=["contraindicates"],
)

Custom types appear after the built-ins in content_types and relation_types. Built-ins always come first.

Additive only, and why

Sending {"content_types": []} clears your custom content types and leaves all nine built-ins in force.

The two vocabularies are independent

Omit a field and that vocabulary is left untouched, so adding a content type cannot accidentally wipe your relation types. Sending neither field is a 422 — an empty request would silently do nothing while returning success.

Naming rules

Rule	Contract
Characters	Lowercase letters, digits and underscores.
First character	Must be a letter.
Length	1–32 characters.
Ceiling	32 custom entries per vocabulary.
Duplicates	Listing the same name twice in one request is rejected.
Restating a built-in	Accepted and quietly ignored — it is already true, so nothing is stored.

Names end up in JSON keys, log fields, metadata values and LLM prompts, which is why they are restricted to a plain identifier.

What a custom type changes

The extractor is told about custom content types and may assign them.
A types filter on recall accepts them instead of silently discarding the name.
A custom relation type becomes valid on Create Relation; before you add it, that request is rejected with a 400 listing the accepted values.

Errors and next action

A 400 names the offending entry: a malformed name, a duplicate, or more than 32 entries. A 400 also results when the caller has no resolvable organization — an ontology has nowhere to live without one, and returning success while writing nothing would be worse.

Safety notes

Create Relation

Use a custom relation type.

Recall Context

Filter recall by content type.

Memory API Overview

See every memory operation.

← Previous

Knowledge Stats

Memory History