Security & Infrastructure
Enterprise-grade security,
built in from day one
Willo runs on modern, fully-managed cloud infrastructure. Every layer is designed for isolation, auditability, and scale. Security is architecture, not an afterthought.
Infrastructure
The Stack
A lean, horizontally scalable foundation built on enterprise-grade, SOC 2-certified providers. No unnecessary complexity. Every choice is deliberate.
Python Backend
Async, high-throughput core powering the orchestration engine and all API surfaces.
Next.js / React
Fast, server-rendered web application with a responsive, accessible component system.
PostgreSQL
Managed relational database with row-level security enforced at the data layer.
Frontier AI
LLM-agnostic architecture: switch to the best model and best rate at any time.
Managed Cloud
SOC 2-certified providers. Horizontally scalable; minimal operational overhead.
Per-Customer Infra
Each business gets its own dedicated website, database, repo, analytics, and email.
Frontend
Next.js / React
Astro sites
Astro sites
→
Backend
Async Python
API + orchestration
API + orchestration
→
Data Layer
PostgreSQL
Row-level isolation
Row-level isolation
→
AI Layer
LLM-agnostic
Frontier models
Frontier models
Platform Security
Security is architecture
Not bolted on. Every surface (auth, permissions, storage, logging) is designed to the same standard.
Authentication
Passwordless and standards-based. Cryptographically signed JWTs validated on every request. No customer passwords are ever stored.
Authorization
Centralized, deny-by-default. A role-based permission model (Owner / Admin / Editor / Viewer) enforced consistently across the entire platform.
Encryption
All data encrypted in transit (TLS) and at rest. Sensitive credentials are additionally encrypted with authenticated encryption and never exposed through the application.
Auditability
Every action is logged with a traceable request ID and attributed to the acting user, giving a complete audit trail across the platform.
Instance Security
Every customer is an island
Data isolation isn't a feature flag. It's enforced at the database level, beneath application code.
Database-level isolation that fails closed.
Customer data is separated using PostgreSQL Row-Level Security, enforced by a restricted database role. Isolation is guaranteed at the data layer itself, not in application code. If a request can't be positively attributed to a tenant, it returns nothing rather than risk exposure.
Database-level RLS
PostgreSQL Row-Level Security enforced by a restricted role. The system fails closed: ambiguous requests return nothing.
Dedicated infrastructure
Tiers with full stack capabilities run on fully separate databases, true physical data separation for customers who need it.
Encrypted, scoped credentials
Any keys or secrets a customer connects are individually encrypted and scoped to that business alone, inaccessible to any other tenant.
AI-native platform
Secrets never touch the AI
We're an AI-native platform — so how we handle customer data inside AI workflows is a first-class design concern, not a compliance footnote.
Out-of-band credential flow
When the AI agent needs a customer's API key or credential, it's supplied through a secure, out-of-band flow. The value never enters the AI's context, prompts, or chat history.
Automatic redaction
If a user accidentally pastes a credential into chat, the platform detects and redacts it before the AI ever sees it, and moves it into the encrypted vault automatically.
Sanitized logging
Sensitive values are stripped from all logs and internal records automatically. No secrets at rest in observability tooling.
No training on customer data
Business data passed through the AI API is not used to train models. Customer prompts and outputs stay private.
How a secret flows →
User input
Pastes or submits a credential
→
Auto-detect
Platform intercepts before AI context
→
Redact
Stripped from chat, logs & history
→
Vault
Stored encrypted, scoped to business
→
AI gets reference
Token only — never the raw value
LLM-agnostic architecture. Willo is not locked to any single model provider. We can switch to the best model and the best rate at any time, giving us flexibility as the AI landscape evolves.
Architected to enterprise standards, with formal certification on our roadmap. Privacy-conscious design throughout, built for the regulatory landscape founders and operators operate in.