Security Architecture

Security Architecture.

Configurable security per request. Self-hosted deployment. Sandboxing on a spectrum. Strict ACL. 100% auditable operations.

Configurable Security Architecture

YOUR RULES.
PER REQUEST.

Every request can be configured with its own security posture — from open access to full network isolation. Self-host for maximum control, or use cloud with per-request policies.

Your Environment

CRM / Email / Tools

Your existing business stack

Your Database

SQLite / Postgres — you own it

Alacritous Cloud

Orchestration Engine

Skills, routing, audit trail

LLM Gateway

Claude / GPT / Gemini

Per-Request Policy

Each call scoped to your rules

Policy-controlled

Configurable Per-Request

Every request can have its own security policy — from fully open to fully isolated. You define the boundaries.

Self-Hosted Option

Deploy the full engine on your infrastructure. Your network, your database, your LLM. Nothing leaves unless you allow it.

BYO LLM

Self-hosted mode supports your own model endpoint. Your prompts stay on your hardware.

Access Control + Sandboxing

EVERY ACTION
PERMISSIONED.

Configurable sandboxing controls what agents can do. Role-based ACL controls what users may do. Every denial is logged.

Sandboxing Level
LooseFully Sandboxed

Strict

Internet for LLM only. No file writes, no third-party APIs.

Internet Access

Enabled

File Scope

Blocked

External APIs

Blocked

Data Export

Blocked

acl.config.yaml

# Role: Admin

-resource:skills.*action:read, write, executeALLOW
-resource:audit.logsaction:read, exportALLOW
-resource:acl.rolesaction:manageALLOW
-resource:data.*action:read, write, deleteALLOW
Pseudo-Deterministic Output

PREDICTABLE
BY DESIGN.

Skills are versioned SOPs. Same input + same Skill version = predictable, auditable output. Every version is hash-pinned like a git commit.

Skill Version:
Input
1> /invoice AcmeCorp
2
3client: AcmeCorp
4billing_info:
5 plan: Enterprise
6 rate: $4,500/mo
7 items: 3 line items
8 tax_region: US-CA
Skill (v2.3)
1name: invoice-generator
2version: 2.3.0
3trigger: /invoice {client}
4
5steps:
6 - fetch: client.billing_info
7 - calculate: line_items × rates
8 - format: pdf_template_v2
9 - deliver: email + slack
10
11output_schema:
12 invoice_id: string
13 total: number
14 status: "draft" | "sent"
a4f29c1immutable skill hash
Output
1invoice_id: INV-2026-0042
2total: $14,850.00
3status: "sent"
4format: pdf_template_v2
5delivered_to:
6 - acme@corp.com
7 - #billing (Slack)

Same Input + Same Skill Version = Predictable Output

Because Skills are versioned and hash-pinned, you can reproduce any past output by re-running the same input against the same Skill version. This gives auditors, regulators, and your team full confidence in what the AI did — and why.

Want to see how these controls work with your infrastructure?

Schedule a Security Walkthrough
Immutable Decision Log active
Compliance Valid
Grc monitor active

DB: SQLite/decisions_trail_v4

Governed Autonomy

AUDITABLE
REASONING.

In 2026, transparency is a compliance requirement. Alacritous logs every decision and rule-change in an immutable audit trail.

When regulatory rules change (like the CTA or NY LLC Act), the OS updates its "Skills" and updates the `decisions` table in your database, providing a perfect trail for regulators.

Regulatory Change Management

"It updates the decisions table in the database, logging exactly how the business's logic was modified to remain compliant."

Strategic Sovereignty

Total Ownership.
Zero Lock-in.

Unlike legacy SaaS that traps your data in walled gardens, Alacritous is built for enterprise sovereignty. You own the stack; we manage the intelligence.

Your Data, Your Control

Alacritous doesn't ingest your data into a central silo. You maintain and own your business database entirely. We simply orchestrate.

Cloud or Self-Hosted

Run Alacritous on your preferred cloud (AWS/Azure/GCP) or deploy on-premise. Own the environment while we provide the managed orchestration layer.

HIPAA-Compliant Clean Rooms

For regulated industries, we partner with proxiML to provide cryptographically secure AI Clean Rooms—zero data exposure, even to us.

AI Operations Role Onboarding

We help you hire and train an internal AI Operations lead. 26% of enterprises now have a CAIO—we prepare your team for this transition.

Governance Certificate // 2026
Database OwnershipClient Verified
Deployment ModelCloud / Self-Hosted
AI Operations TrainingRole Onboarding
Enterprise Security Partner
SOC 2

Powered by proxiML® AI Clean Rooms for HIPAA-compliant, cryptographically secure deployments.

Zero Data ExposureOn-Premise ReadyHIPAA

100% Non-Custodial Intelligence

Compliance Scenarios

REGULATORY
READINESS.

Simulate compliance scenarios across jurisdictions. See how Alacritous maps to real regulatory requirements.

Risk Analysis active

Corporate Transparency Act

High Risk

"Failure to report Beneficial Ownership Information (BOI) carries civil penalties of up to $500 per day."

Alacritous Mitigation

Alacritous autonomously monitors FinCEN rule changes and drafts required disclosures for human approval.

Governed Autonomy Active
Secure Compliance
Security Fundamentals

BUILT-IN
SECURITY.

Encryption at Rest

AES-256 encryption for all stored data. Your database, your keys.

Encryption in Transit

TLS 1.3 on every connection. No plaintext, ever.

SSO & SAML

Enterprise SSO integration. SAML 2.0, OAuth 2.0, OpenID Connect.

RBAC

Role-based access control at every layer — users, agents, Skills, data.

API Key Rotation

Automated key rotation with zero-downtime credential cycling.

Rate Limiting

Per-endpoint rate limiting prevents abuse and resource exhaustion.

Dependency Scanning

Automated CVE scanning on every dependency, every build.

Incident Response

Documented incident response process with defined SLAs and escalation paths.

FAQ

SECURITY
QUESTIONS.

Can Alacritous be self-hosted on our intranet?

Yes. Alacritous supports full self-hosted deployment on your own infrastructure — AWS, Azure, GCP, or bare metal. The orchestration engine, database, and all Skills run entirely within your network perimeter. No data leaves your environment unless you configure outbound integrations.

How configurable is the network security?

Fully configurable per request. You can lock down individual requests to have no external access, allow scoped outbound calls to specific APIs, or run fully open — it's your policy. In self-hosted mode, you control the entire network perimeter. Every configuration is logged and auditable.

What LLM does Alacritous use? Can we bring our own?

In cloud-hosted mode, Alacritous routes through major LLM providers (Claude, GPT, Gemini) via secure API calls. In self-hosted mode, you can bring your own LLM endpoint — including private models running on your infrastructure. Your prompts never leave your network.

How does Alacritous prevent data exfiltration?

Through layered controls: role-based access control (RBAC) restricts what each user and agent can access, configurable sandboxing limits agent capabilities (network access, file scope, data export), and every action — including denials — is logged to an immutable audit trail.

Can we control how sandboxed our AI agents are?

Absolutely. Sandboxing is configurable on a spectrum from 'Loose' (full access for trusted internal automation) to 'Fully Sandboxed' (complete isolation with no network, file, or export capabilities). Each capability — internet access, file scope, external APIs, data export — can be independently controlled.

Is every AI action logged and auditable?

Yes, 100%. Every agent action, every decision, every permission check, and every denial is logged in an immutable audit trail stored in your database. The trail includes timestamps, the Skill version used, the user who initiated the action, and the full reasoning chain. This satisfies requirements for SOC 2, EU AI Act, and HIPAA audit readiness.

How deterministic are AI agent outputs?

Alacritous achieves pseudo-deterministic output through versioned Skills. Each Skill is a hash-pinned SOP — same input plus same Skill version produces predictable, reproducible output. When a Skill is updated, the version increments and the previous version remains available for audit reproduction.

What standard security practices does Alacritous follow?

Alacritous implements encryption at rest and in transit (AES-256 / TLS 1.3), SSO and SAML integration, role-based access control, API key rotation, rate limiting on all endpoints, automated dependency scanning, and a documented incident response process. SOC 2 Type II readiness is built in.

SECURE YOUR
OPERATIONS.

"Your rules, per request. Full auditability. Your infrastructure, your models, your data."

30 minutes. We'll walk through your compliance requirements. No commitment.