Stop dangerous memory drift before it cascades through your agent network.
Try it free: sg_demo_playground
In multi-agent systems, a single corrupted or stale memory entry can propagate through dozens of agents before anyone notices. By the time the damage surfaces, it's too late.
One agent's memory becomes another's fact. Stale data compounds at every hop.
No warnings until an irreversible action is taken. The damage is invisible until it's too late.
downstream_count tracks how far each entry spreads across your fleet.
Agent A → memory → [SGRAAL PREFLIGHT] → Agent B
↓
USE_MEMORY / WARN / ASK_USER / BLOCK
↓
propagation_trace + blast_radius
Full chain visualization — every hop, every agent.
POST /v1/propagation/trace
downstream_count × action_risk = total impact score.
SUCCESS / PARTIAL / FAILED — how far has it spread?
BLOCK before the cascade completes. Surgical containment.
614/614
Corpus cases — 0 false negatives
F1 = 1.000
4 rounds · joint benchmark with Grok
<180ms
Avg latency · Grok dual-stack
<2%
Blast radius · propagation contained
from sgraal import SgraalClient client = SgraalClient("sg_demo_playground") result = client.preflight( memory_state=[...], domain="fintech", action_type="irreversible" ) trace = client.propagation_trace( agent_id="agent_001", memory_state=[...], domain="fintech" ) print(trace["affected_agents"]) # 12 downstream agents print(trace["containment"]) # "SUCCESS"
One poisoned memory. Eleven agents in the network. See exactly how far it spreads — with and without Sgraal.