Multi-Tenancy

How it works

MemWire has a built-in four-level isolation hierarchy. Every memory, graph node, graph edge, and knowledge chunk is tagged with up to four identifiers:

org_id  →  workspace_id  →  app_id  →  user_id

All queries like recall, search, and knowledge are automatically scoped to the identifiers you pass. No cross-tenant data leaks are possible because every vector storage filter and SQL index is keyed on this hierarchy.

Levels explained

Level	Identifier	Typical use
Organisation	`org_id`	Top-level billing or company boundary
Workspace	`workspace_id`	Team, project, or environment (prod vs staging)
Application	`app_id`	Individual product or service
User	`user_id`	End-user required on every call

You only need to use the levels that make sense for your architecture. user_id is always required; the others default to None.

Code example

from memwire import MemWire, MemWireConfig

# Set org_id at config level — applies to all operations
config = MemWireConfig(
    qdrant_path="./memwire_data",
    org_id="acme_corp",
)
memory = MemWire(config=config)

# Add memory scoped to a workspace + app + user
memory.add(
    user_id="alice",
    workspace_id="team_design",
    app_id="design_assistant",
    messages=[{"role": "user", "content": "I prefer minimalist UI patterns."}],
)

# Recall is automatically scoped — alice in design_assistant only
result = memory.recall(
    "What are my design preferences?",
    user_id="alice",
    workspace_id="team_design",
    app_id="design_assistant",
)
print(result.formatted)

# Bob in a different workspace never sees alice's memories
result_bob = memory.recall(
    "What are my design preferences?",
    user_id="bob",
    workspace_id="team_engineering",
    app_id="code_assistant",
)
print(result_bob.formatted)  # empty — no cross-contamination

Agent isolation

Pass agent_id to further scope memories to a specific AI agent within a user session:

memory.add(
    user_id="alice",
    agent_id="planner_agent",
    messages=[{"role": "assistant", "content": "I scheduled the meeting for Friday."}],
)

result = memory.recall(
    "When is the meeting?",
    user_id="alice",
    agent_id="planner_agent",
)

Configuration reference

Parameter	Default	Description
`org_id`	`"default"`	Organisation identifier set at config level.
`database_url`	`None`	SQLAlchemy URL. Defaults to `sqlite:///memwire_{org_id}.db`.

Getting started

Features

Configuration

How it works

Levels explained

Code example

Agent isolation

Configuration reference

​How it works

​Levels explained

​Code example

​Agent isolation

​Configuration reference

How it works

Levels explained

Code example

Agent isolation

Configuration reference