And you won't see it happen.
Memory poisoning, prompt injection, and tampering are blocked before they reach your agent.
Get API Key arrow_forwardWithout Sgraal
With Sgraal
The write firewall intercepts incoming memory. Sleeper patterns, injection signatures, and source anomalies are flagged immediately.
Poisoning detection, hallucination risk, tamper verification (Merkle hash), and circuit breaker on repeated high-risk patterns.
Memory Forensics · Black Box Capsule (SHA256) · ZK Validation. Every threat is traceable. Every decision is defensible.
poisoning_suspected: true — 3-signal combination. hallucination_risk: high — cyclic contradiction in memory graph. Circuit breaker: 5× omega > 80 → HTTP 429 Safety Block.
Your agent acts on corrupted data — silently, repeatedly, without you knowing.
Write firewall: blocks poisoned data before storage — prevention not detection. Sleeper detector: proactive scan for dormant trigger conditions.
Never store poisoned memory. Ever.
Most tools detect after storage. Sgraal blocks before it ever enters memory.
Memory Forensics: full incident trace. Black Box Recorder: tamper-evident capsule for every BLOCK. ZK Validation: omega score computed without content leaving your system.
Every threat is provable in court. Every block is auditable.
Red Team as a Service: injection, poisoning, replay, drift, tamper, sleeper. Memory Readiness Grade: A–F graded assessment.
Know your resilience score before attackers find out for you.
Block poisoned memory at write time. Detect tampering at read time.
# Store with write firewall — blocks before storage
result = client.store(
content="User account balance: $50,000",
agent_id="agent-payments",
domain="fintech"
)
if result.write_firewall_triggered:
print(result.block_reason) # SLEEPER_PATTERN_DETECTED
print(result.threat_score) # 0.94
# Preflight scan — detect poisoning at read time
preflight = client.preflight(
memory_state=[{"id": "mem_001", "content": "...",
"type": "semantic", "timestamp_age_days": 1}],
action_type="irreversible",
domain="fintech"
)
print(preflight.poisoning_suspected) # True
print(preflight.recommended_action) # BLOCK
When one agent detects an attack, Sgraal extracts the attack signature and immunizes every agent in your fleet — automatically, in under a second. No manual intervention. No configuration. One agent attacked, all agents protected.
Powered by cryptographically signed vaccine signatures with automatic expiry.