Build the decision plane for AI agents
Memrail is building SOMA AMI™ — the decision plane for AI agents. Deterministic context assembly, policy gates, and auditable traces. We're hiring engineers who want to make agent behavior reproducible in production.
4
Open roles
Miami
Florida
# Deterministic decisioning + traceability
async with AsyncAMIClient() as client:
resp = await client.invoke(
atoms=[state('customer.tier', 'vip'), state('ticket.priority', 'high'), tag('intent', 'cancel')],
options=InvokeOptions(dry_run=True, explain=True),
trace=TraceOptions(enable=True)
)
# resp.selected → actions • resp.trace → why
async with AsyncAMIClient() as client:
resp = await client.invoke(
atoms=[state('customer.tier', 'vip'), state('ticket.priority', 'high'), tag('intent', 'cancel')],
options=InvokeOptions(dry_run=True, explain=True),
trace=TraceOptions(enable=True)
)
# resp.selected → actions • resp.trace → why
What We're Building
Decision Plane Runtime
EMU runtime, trigger reachability, policy compilation, deterministic activation, idempotency.
SDK & Integrations
Python-first SDK plus integrations for LangGraph, n8n, LlamaIndex, and custom orchestrators.
Tracing & Replay
Decision traces, diffable replays, and debugging tools for cognitive systems in production.
Open Positions
Memrail is part of the Cadenzai family — research (Cadenzai), infrastructure (Memrail), and applied practice (CognitionEngineers). Roles collaborate across entities.