ai-memory-layer-client

Python client for the memory-layer HTTP service.

Install

pip install ai-memory-layer-client

For local development and tests:

pip install -e '.[dev]'
pytest

Quick Start

from memory_layer_client import MemoryClient, MemoryRuntimeClient, MemoryScope

client = MemoryClient(
    "http://localhost:3100",
    api_key="dev-key",
    default_scope=MemoryScope(
        tenant_id="acme",
        system_id="python-worker",
        collaboration_id="release-42",
        scope_id="run-42",
    ),
)
runtime = MemoryRuntimeClient(client)

result = runtime.run_turn(
    "Remember that deployments must stay blue-green.",
    lambda prepared: "Stored. I will keep rollout constraints in memory.",
)
print(result.prepared.prompt)

Async Client

import asyncio
from memory_layer_client import AsyncMemoryClient, AsyncMemoryRuntimeClient

async def main() -> None:
    client = AsyncMemoryClient("http://localhost:3100", api_key="dev-key")
    runtime = AsyncMemoryRuntimeClient(client)
    result = await runtime.run_turn(
        "Track migration status.",
        lambda prepared: "I will keep the migration status in memory.",
    )
    health = await client.health()
    print(result.prepared.context.token_estimate, health.active_turn_count)
    await client.aclose()

asyncio.run(main())

Scope Resolution

Each request can inherit a default scope and override it per call. This matches the multi-tenant routing model used by the HTTP server.

MemoryScope also supports collaboration_id, so Python clients can participate in shared multi-agent workspaces the same way the TypeScript server and SDK do.

Hosted Runtime And Changes

from memory_layer_client import MemoryClient, MemoryRuntimeClient

client = MemoryClient("http://localhost:3100")
runtime = MemoryRuntimeClient(client)

knowledge = client.list_knowledge(limit=20)
detail = client.inspect_knowledge(knowledge.items[0]["id"])
changes = client.poll_changes("2026-03-01T00:00:00Z", scope_level="workspace")
cross_scope = client.search_cross_scope("rollback", scope_level="workspace")
prepared = runtime.before_model_call("What changed in the shared workspace?")
print(prepared.prompt)
for event in client.stream_events(event_types=["knowledge_change"], scope_level="workspace"):
    print(event.type, event.meta)
    break

CLI

memory-layer-client --base-url http://localhost:3100 health
memory-layer-client --base-url http://localhost:3100 search "rollback checklist"
memory-layer-client --base-url http://localhost:3100 inspect-knowledge --limit 10
memory-layer-client --base-url http://localhost:3100 run-reverification --admin-key secret