Memside Python SDK

Python client for Memside AI continuity, portable context, and Library Templates.

Documentation: Python SDK

Install

pip install memside

Usage

from memside import MemsideClient

client = MemsideClient(api_key="mem_sk_example_key")

startup = client.context_startup()
print(startup)

Create a memory with the public type and text fields:

memory = client.memories_create(
    {
        "type": "note",
        "text": "Packing checklist\nBring a charger and a reusable bottle.",
        "sensitivity": "private",
    }
)

Use the current version when updating a previously read Memory:

updated = client.memories_update(
    memory["id"],
    {
        "text": "Packing checklist\nBring a charger and a reusable bottle.",
        "expected_version": memory["version"],
    },
)

Create a Subject and link a non-secret memory:

subject = client.subjects_create(
    {
        "name": "Weekend reading",
        "subject_type": "topic",
        "aliases": ["Reading list"],
    }
)

client.subjects_link_memory(subject["id"], memory["id"])
context = client.subjects_get_context(subject["id"])

Read eligible Facts and propose a source-backed addition for user review:

facts = client.subjects_list_facts(subject["id"])

suggestion = client.subjects_suggest_fact(
    subject["id"],
    {
        "suggestion_type": "add_fact",
        "proposed_fact_type": "preference",
        "proposed_fact_text": (
            "Prefers printed books for long-form reading."
        ),
        "source_memory_id": memory["id"],
        "idempotency_key": "reading-preference-v1",
    },
)

Fact Suggestions remain pending until a signed-in user reviews them. The SDK does not expose direct Fact mutation or suggestion review.

Read a bounded Context Map and pending Memory Insights:

context_map = client.memories_get_context_map(memory["id"])
history = client.memories_get_revisions(memory["id"])
insights = client.subjects_list_memory_insights(
    subject["id"],
    limit=25,
)

Insight refresh, review, and undo remain signed-in application actions.

Prepare Subject deletion without changing data:

preview = client.subjects_prepare_delete(subject["id"])
print(preview["required_confirmation"])

The SDK does not generate or submit Subject deletion confirmation automatically.

Deletion requires the resource-specific confirmation returned by the calling application workflow:

client.memories_delete(
    memory["id"],
    f"CONFIRM_DELETE_{memory['id']}",
)

Search public Library Templates:

templates = client.library_search_templates(
    query="project planning",
    limit=5,
)

Read an owned draft before making a revision-protected update:

draft = client.library_read_template(
    "template-id",
    {"file_paths": ["README.md"]},
)

client.library_write_draft(
    "template-id",
    {
        "expected_fingerprint": draft["content_fingerprint"],
        "idempotency_key": "planning-template-v1",
        "files": [
            {
                "path": "README.md",
                "content": "# Project planning\n\nA simple planning checklist.",
                "target": "reusable_template",
            }
        ],
    },
)

Template publishing remains a separate workflow action and requires the permissions and confirmations defined by the public contract.

The API key can also be set through the environment:

MEMSIDE_API_KEY=mem_sk_example_key

Supported API Areas

This package wraps public Memside API-key routes:

• startup context
• resume context
• workspace profile
• memory listing
• memory search
• memory fetch
• memory revision history
• bounded exact-ID memory batch reads
• bounded memory Context Maps
• memory create
• memory update
• memory delete, when allowed by the API
• Subject listing, creation, retrieval, and update
• Subject-memory listing, linking, and unlinking
• bounded Subject Context
• eligible Subject Fact reads
• source-backed Fact Suggestions for signed-in-user review
• pending-only Memory Insight reads
• guarded Subject deletion preparation and confirmation
• public Library Template search
• creator-owned Template status and draft reads
• revision-protected Template draft replacement
• reviewed Template publication and visibility workflows

This package does not include private Memside application source, account/session internals, billing internals, admin routes, database details, or MCP server implementation.

API Reference

See the curated OpenAPI document for public request and response fields.

Public request and response shapes are exported as lightweight TypedDict definitions for editors and type checkers. Failed requests raise MemsideError, including status, code, retryable, retry_after, request_id, and safe details.

For 429 responses, wait for retry_after before retrying a read. The SDK does not automatically retry writes.

Eligible responses contain attachment metadata only. Raw attachment data, private storage locations, and signed download URLs are not returned through the public API-key contract.

Compatibility

The package uses additive public methods and typed fields where possible. Breaking changes require a documented migration path. The client accepts the current public error envelope and documented legacy detail responses during the V2 compatibility period.

Requirements

Python 3.9 or newer.