Skip to content
ONE3.ai

Platform

The platform under every One3 app

One kernel, one data model, one identity and one AI platform. Every app, Action and Digital Employee runs on the same governed foundation.

Why One3

One3.ai is planned as two parts that reinforce each other: an AI Business OS built on one kernel, and a Digital Employee Economy where clients rent governed AI agents. This page explains the vision, who One3 serves first, the rules it will not break, and what it will not do.

Explore

Kernel & Actions

Layer 0, the Kernel, is the only thing every One3 layer depends on. It defines 14 primitives and one Action contract that the UI, API, workflows, MCP and Digital Employees all call. Packs bundle these primitives into the 200 planned apps.

Explore

AI Platform

The One3 AI platform combines a Model Hub and router, a durable agent runtime, autonomy levels A0–A4, a Context Engine with permission-aware retrieval, layered guardrails, an eval system built before the agents, and AI FinOps that reports cost per business outcome.

Explore

Enterprise Brain

One3's Enterprise Brain is a planned layer of reasoning services over the kernel: a Signal Hub, World Model, Goal Stack, Decision Engine, Critic, Governor, Executor and Calibration Engine, running autonomous loops at four clock speeds under human authority.

Explore

One3 Models

One3's model strategy moves each task class between four tiers on measured evidence, plans a family of specialist One3 models, and improves them through a governed pipeline of six loops, business gyms and strict evaluation, with humans approving every promotion.

Explore

Identity & Access

One3's identity layer plans one global login with a membership per tenant, passkeys and enterprise SSO/SCIM, RBAC for what you can do, relationship-based access for which records, field-level security, separation of duties, agent permissions bounded by their owners, and a compiled entitlement engine.

Explore

Data Architecture

One3's data architecture puts all apps on one shared model in PostgreSQL. A shared party record holds identity, contact points and consent. Each entity has one owner module. Changes publish events through an outbox. Custom fields need no migration. Erasure, retention, legal hold and backup are designed in from the start.

Explore

Security & Trust

One3's security architecture plans four tenant isolation tiers, two-layer isolation for rented Digital Employees, layered identity and authorisation, agent safety controls, encryption with per-tenant keys, DPDP-aware erasure, a hash-chained audit log, marketplace trust levels and strict public claim rules.

Explore

Integrations

One3's integration plan has five coverage lanes, one connector registry and a Connector Health Center that agents must check before acting on connector data. MCP, OpenAPI, CloudEvents and other open standards sit behind One3 abstractions, and India Stack rails are planned through licensed partners.

Explore

Migration

One3's Migration Factory is planned to connect to tools such as HubSpot, GoHighLevel, Zoho CRM, Freshdesk, Pipedrive and spreadsheets. AI suggests field mappings, a person approves them, a dry run checks everything without writing, and a reconciliation report compares source and One3 before sign-off. Every migrated batch can be rolled back.

Explore

Performance

One3's performance architecture assigns every Action, view and API a latency class with p50/p95/p99 targets, keeps the hot path to one or two database trips, caches per tenant with versioned keys, separates interactive and batch work, and enforces multi-tenant fairness.

Explore

Technology

One3's planned stack: TypeScript-first product code on NestJS with the Fastify adapter, PostgreSQL 18 with Drizzle ORM and row-level security, Temporal and NATS JetStream for workflows and events, Next.js and Expo on the front end, and Python with FastAPI only for AI. A complexity budget caps production at six core infrastructure systems in W0–W1.

Explore

Open Source

One3's plan evaluates 100 open-source candidates against ten written decision rules. Permissive and unmodified MPL-2.0 engines are the default, a short list of copyleft components runs only as isolated services, source-available licences are never used for anything customer-facing, and every engine sits behind a replaceable One3 interface with a Fact Register row.

Explore

Deployment & Edge

One3's deployment plan moves in stages from a build machine to cloud cells in India, then to multi-region cells, One3 Box appliances and edge nodes. A deployment profile toggles features and model routes, and offline sync uses the event log so work continues without a network.

Explore

See how One3 would run your business

Tell us about your team and the outcomes you want. We will walk you through the apps, the Digital Employees and the plan for your industry.