/ Contact · we read every inquiry
Talk to 
.
Demos, partnerships, government RFPs, technical questions. A person reads every form. You hear from someone — not a queue.
Security is not a feature — it is the architecture. Every request authenticated, every action logged, every byte in your custody. No exceptions.
Complete transparency into how CASTLE protects your data, your models, and your infrastructure.
Every request passes through authenticated, authorized, and audited layers before reaching inference. No exceptions.
Every request authenticated. Every action logged. Every byte in your custody.
Shield validates identity via API key or SSO. Tower enforces role-based access. Echo processes inference on-premises. Atlas records an immutable audit trail.
Our compliance posture is designed for Canadian government and regulated enterprise requirements.
| Standard | Status | Summary | |
|---|---|---|---|
| PIPEDA | Compliant | Personal information stays on Canadian infrastructure | ▶ |
| ITSG-33 | Aligned | 11 baseline security controls documented | ▶ |
| FIPS 140-2 | In Progress | Level 1 software certification underway | ▶ |
| Protected B | Assessment Planned | PBMM readiness documented | ▶ |
| SOC 2 Type II | Planned | Audit engagement scheduled Q3 2026 | ▶ |
| Canadian Data Residency | Compliant | All data on Canadian-owned hardware | ▶ |
Zero external API calls. Inference runs on your hardware. Models execute locally with no network dependency for core operations.
Data is never transmitted to AXE or any third party. Your prompts, responses, training data, and model weights remain exclusively yours.
Canadian-owned company. Canadian infrastructure. No foreign intelligence access. Not subject to US CLOUD Act, FISA, or Patriot Act compulsion.
All data is encrypted at rest, in transit, and during key management operations using industry-standard cryptographic primitives.
All data stored in Atlas tables is encrypted using AES-256 via pgcrypto. Database-level encryption ensures data remains protected even if physical media is compromised.
TLS 1.3 enforced on all public-facing endpoints. Inter-node communication within fleet deployments uses WireGuard tunnels with 256-bit keys.
API keys derived via PBKDF2-SHA256 with high iteration counts. Hardware RNG used where available for entropy. Key rotation supported without service interruption.
Our team includes active security researchers. CASTLE is tested against the same standards we use to audit client infrastructure.
Automated weekly security scans run across all CASTLE nodes. Vulnerability detection covers network exposure, service configuration, and dependency analysis.
The castle-harden.sh audit script is available for clients to independently verify the security posture of their deployment. No black boxes.
All CASTLE API endpoints are tested against the current OWASP Top 10. Injection, broken authentication, security misconfiguration, and all other categories covered.
Prompt injection, data exfiltration, and adversarial input testing performed on all deployed models. Red team exercises run quarterly.
Tamper-evident logging with cryptographic integrity verification. DELETE operations are blocked at the database level.
CREATE TABLE audit_log ( id BIGSERIAL PRIMARY KEY, event_type VARCHAR(64) NOT NULL, -- auth, query, admin, error user_id UUID NOT NULL, ip_address INET NOT NULL, endpoint VARCHAR(256), details JSONB, -- request metadata, model, tokens checksum CHAR(64) NOT NULL, -- SHA-256 of previous row + current created_at TIMESTAMPTZ DEFAULT NOW() ); -- DELETE operations blocked via trigger CREATE RULE no_delete AS ON DELETE TO audit_log DO INSTEAD NOTHING; -- Integrity verification SELECT verify_audit_chain('2026-01-01', '2026-04-07'); -- Returns: OK | row_id of first broken link
SHA-256 hash chain links every row to its predecessor. Tampering breaks the chain and is immediately detectable.
DELETE operations are blocked at the database level via PostgreSQL rules. Audit records are append-only and permanent.
Every API call, authentication event, model query, and administrative action is logged with full request context and IP attribution.
Our incident response plan is tested and documented. We commit to aggressive timelines and transparent disclosure.
Automated monitoring detects anomalies. Sentinel health checks run continuously across all fleet nodes.
Affected systems isolated. Threat vector identified and blocked. Forensic preservation initiated.
Affected clients notified with incident details, scope assessment, and remediation steps.
Root cause analysis published. Preventive controls deployed. Post-incident review shared with stakeholders.
For security concerns or responsible disclosure, contact our security team directly. We acknowledge all reports within 24 hours and follow coordinated disclosure practices.
Download our security whitepaper or schedule a technical review with our security team.
/ Contact · we read every inquiry
.Demos, partnerships, government RFPs, technical questions. A person reads every form. You hear from someone — not a queue.
AXE Concierge · live
Answered by a local model · Canadian iron · zero external API