Shared Memory

Shared memory powers coordination between multiple agents. It sits in src/memorizz/coordination/shared_memory/ and corresponds to MemoryType.SHARED_MEMORY.

Why It Exists

Complex workflows often split responsibilities across researcher, analyst, and writer agents. Shared memory provides a blackboard-like store where agents can exchange artifacts, delegate tasks, and keep track of global progress.

Creating a Session

from memorizz.coordination.shared_memory import SharedMemory

shared = SharedMemory(memory_provider)
session_id = shared.create_shared_session(
    root_agent_id="orchestrator",
    delegate_agent_ids=["researcher", "writer"],
    workflow_id="market-review-2026-08-21",
    user_id="user-42",
    trace_id="trace-01",
)

shared.post_command(
    memory_id=session_id,
    agent_id="orchestrator",
    command_id="research-1",
    target_agent_id="researcher",
    instructions="Collect primary sources for the market review.",
)

Each session keeps:

• Participants and roles
• Messages and artifacts exchanged between agents
• Links to the originating episodic/semantic records for traceability

workflow_id and user_id are part of the isolation contract. Do not use a global shared-memory session for unrelated users or workflows. Pass tenant and trace context through orchestration, and validate participant ownership before posting or reading artifacts.

Patterns

• Orchestrator + delegate setups (research, summarization, QA)
• Human-in-the-loop review queues where both agents and operators inspect shared state
• Multi-modal agents handing off voice, vision, or text data through a common buffer

Shared memory complements the per-agent stores so everyone observes the same document trail without duplicating data.