Skip to main content
Every API key belongs to a workspace. Memory is fully isolated between workspaces — no data can cross workspace boundaries.

​
How it works

The server extracts the workspace from the API key on every request. All memory operations are scoped to that workspace automatically. 
# Agent A — workspace "acme"
# PLYRA_API_KEY=plm_live_acme...
async with Memory(agent_id="agent-1") as mem:
    await mem.remember("acme internal strategy")

# Agent B — workspace "other"
# PLYRA_API_KEY=plm_live_other...
async with Memory(agent_id="agent-1") as mem:
    ctx = await mem.context_for("acme internal strategy")
    # → returns nothing. Workspace isolation enforced server-side.

​
Namespacing within a workspace

Within a workspace, memory is further namespaced by agent_id: 
# Support agent — only sees its own memory
async with Memory(agent_id="support-agent") as mem:
    await mem.remember("user prefers dark mode")

# Sales agent — different namespace, cannot see support memory
async with Memory(agent_id="sales-agent") as mem:
    ctx = await mem.context_for("user preferences")
    # Returns nothing — different agent_id namespace

​
One key per use case

Create separate API keys for:
  • Each environment (dev / staging / prod)
  • Each team or customer
  • Each agent type, if strict isolation is required
Revoke a single key to cut off one tenant without affecting others.