python

Memind Python Client

Official Python client for the Memind memory engine API.

Installation

pip install memind

Synchronous Usage

from memind import MemindClient, Message, Strategy
from memind.types import ConversationContent

with MemindClient(base_url="http://localhost:8080") as client:
    health = client.health()

    response = client.memory.extract(
        user_id="user-1",
        agent_id="agent-1",
        raw_content=ConversationContent(messages=[Message.user("Remember that I prefer concise answers.")]),
    )
    print(response.status)

    result = client.memory.retrieve(
        user_id="user-1",
        agent_id="agent-1",
        query="What does the user like?",
        strategy=Strategy.SIMPLE,
        trace=True,
    )

memory.extract() uses Memind's synchronous extraction endpoint and returns ExtractMemoryResponse. Treat only status == "SUCCESS" as safe to clear caller-owned retry payloads; PARTIAL_SUCCESS is surfaced so applications can keep or re-enqueue the original payload.

Agent Timeline Raw Content

Coding-agent integrations can submit tool and command activity as agent_timeline raw data:

response = client.memory.extract_agent_timeline(
    user_id="local__alice",
    agent_id="claude-code__project_hash",
    source_client="claude-code",
    timeline={
        "sourceClient": "claude-code",
        "sessionId": "session-123",
        "agentTurnId": "session-123-agent-turn-1-1",
        "timelineId": "session-123-agent-1-2",
        "events": [
            {
                "eventId": "event-id",
                "seq": 1,
                "kind": "command",
                "toolName": "Bash",
                "command": "npm test payment",
                "status": "failed",
                "exitCode": 1,
                "output": '{"stdout": "rounding mismatch"}',
            }
        ],
    },
)

The helper sends rawContent.type = "agent_timeline" through the same synchronous extraction endpoint.

Asynchronous Usage

from memind import AsyncMemindClient, Strategy

async with AsyncMemindClient(base_url="http://localhost:8080") as client:
    result = await client.memory.retrieve(
        user_id="user-1",
        agent_id="agent-1",
        query="What does the user like?",
        strategy=Strategy.DEEP,
    )

The async resource also provides await client.memory.extract_agent_timeline(...) with the same arguments as the synchronous helper.

Configuration

Configuration precedence:

1. Constructor arguments
2. Environment variables: MEMIND_BASE_URL, MEMIND_API_TOKEN
3. Defaults: connect timeout 5s, read timeout 30s, max retries 2

base_url is required. api_token is optional; when omitted no Authorization header is sent.

Development

pip install -e ".[dev]"
ruff format .
ruff check .
mypy src
pytest --cov=memind --cov-fail-under=90 -q
python -m build
twine check dist/*