Platform

The runtime for AI accountability.

Five governance checkpoints. Twenty-seven guardrail categories. One hundred and nine configurable policies. Zero exceptions. Every AI interaction in your enterprise passes through Raidu and leaves behind a cryptographically signed record.

Runtime-native.
Not a dashboard
on the side.
EU AI ACT · ENFORCES 2026-08-02 HIPAA AI RULE · MAY 2026 COLORADO SB 24-205 · JUNE 2026 NYDFS CIRCULAR LETTER NO. 7 SR 11-7 MODEL RISK MANAGEMENT GDPR · DPDPA · PIPL · LGPD · POPIA ISO/IEC 42001 AIMS NIST AI RMF 1.0 OMB M-24-10 EU AI ACT · ENFORCES 2026-08-02 HIPAA AI RULE · MAY 2026 COLORADO SB 24-205 · JUNE 2026 NYDFS CIRCULAR LETTER NO. 7 SR 11-7 MODEL RISK MANAGEMENT GDPR · DPDPA · PIPL · LGPD · POPIA ISO/IEC 42001 AIMS NIST AI RMF 1.0 OMB M-24-10

In the path. Not beside it.

Raidu sits between every human, agent, or workflow and the models and tools they reach for. Nothing bypasses the runtime.

HUMAN (CHAT)
AGENT
WORKFLOW
IDE ASSISTANT
API CLIENT
RAIDU RUNTIME
5 checkpoints
01
02
03
04
05
PII MASK · POLICY · SIGNING · WORM
175 LLMs / 24 PROVIDERS
GMAIL · OUTLOOK · SLACK
JIRA · LINEAR · ASANA
STRIPE · HUBSPOT · ZENDESK
S3 · SNOWFLAKE · BIGQUERY
◆ AUDIT BUS · RSA-4096 SIGNED · HASH CHAINED · WORM 10Y write-only · regulator-verifiable
The runtime

Five checkpoints. Each explainable. Each signed.

CHECKPOINT 01 · USER INPUT · user prompt

PII detected and logged without masking. The AI understands the original request. The governance record is already open.

What fires
detect_entities log_raw attach_context open_execution
Example trace
INPUT → "send contract to jane.doe@acme.com, SSN 412-55-9021, card 4242 4242 4242 1234"
EMAIL US_SSN CREDIT_CARD
<100ms
overhead
99.2%
accuracy
signed
every time
Guardrails

27 categories. 109 policies.

Every guardrail runs inline. Every match produces a plain-English explanation. Every decision is signed.

01
PII Detection INPUT
SSN, credit card, phone, email, address, 60+ entities
02
PHI Detection INPUT
HIPAA-scoped patient identifiers, MRN, diagnosis codes
03
PCI Detection INPUT
Card PAN, CVV, track data, cardholder name
04
Prompt Injection INPUT
Direct, indirect, multi-turn, encoding tricks
05
Jailbreak INPUT
Roleplay, DAN, persona exploits, context escalation
06
Secrets PRE-INF
AWS keys, GitHub tokens, API keys, private keys
07
Topic Policy PRE-INF
Allow / deny lists per team, per model
08
Model Policy PRE-INF
Approved model list · per-use-case routing
09
Tool Policy TOOL
Which agents can call which connectors
10
Scope Policy TOOL
Which records each tool call is allowed to touch
11
Budget Policy TOOL
Per-user, per-agent, per-model spend caps
12
Rate Policy TOOL
Requests per second, burst control
13
Data Exfiltration POST-INF
Outbound secrets, tokens, keys, internal IDs
14
Toxicity POST-INF
Hate, harassment, violence, self-harm
15
Bias POST-INF
Protected class, hiring, lending, healthcare
16
Hallucination POST-INF
Cross-reference against source documents
17
Groundedness POST-INF
Confidence scoring for every factual claim
18
Content Filter POST-INF
Sexual, CSAM, extremism, weaponry
19
Code Safety POST-INF
Malware, license risk, credential leaks in generated code
20
Approval Gates POLICY
Human-in-the-loop for high-risk actions
21
Masked Logging POLICY
Never log raw PII anywhere after ckpt 02
22
Custom POLICY
Drag-and-drop policy builder, versioned
23
Data Residency COMPLIANCE
EU · US · APAC region pinning
24
Retention Policy COMPLIANCE
TTL per data class, redaction on expiry
25
Regulatory Mapping COMPLIANCE
EU AI Act · HIPAA · GDPR · SB 24-205
26
Legal Hold COMPLIANCE
Freeze records on litigation trigger
27
Audit Replay COMPLIANCE
Re-run past executions with new policy
Connector-aware firewall

Gmail sees emails. Stripe does not.

Every connector declares the PII it needs to function. The firewall adapts per connector. The table is generated from the connector manifest, not hand-written. Change a manifest and every execution after that change can prove it used the new policy.

Connector NAME EMAIL PHONE ADDRESS SSN CREDIT_CARD MRN
Gmail
Outlook
Slack
Stripe
Salesforce
Zendesk
Jira
GitHub
HubSpot
S3
ALLOWED DENIED (MASKED) N/A
Latency

Under 10 ms per checkpoint.

Raidu adds governance to the request path without adding latency anyone will feel. The slowest operation, the signature, takes longer than most of the checkpoint stack combined and still finishes in under four milliseconds at p95.

Measured on n2-standard-4 · GCP us-east1 · April 2026
Operationp50p95
PII detect (Presidio) 2.1 4.8
Mask + token map 0.4 1.1
Policy evaluation 0.8 2.4
Connector rule merge 0.6 1.9
RSA-4096 sign 1.4 3.2
Hash-chain append 0.3 0.9
WORM write (async)ASYNC
Total (sync path) p95 14.3 ms · p50 5.6 ms
46 / 46
EU AI Act features
175
Models supported
34
Connectors · 379 tools
24
Model providers
99.2%
PII detection accuracy