Product

Agentic Backend

A backend built from the business definition. Rewrite the definition and the change takes effect, synced to the agents.

The backend for the agent era

Purpose-built for AI-native business

The agent era's business system can't be built as an extension of the old one. We started the design from that shift.

The principles

01

Definition-driven

02

Transparency

03

Security

The value

Hand the business to agents

Inside defined operations and rules, agents run the business end to end.

Changes land the same day

Rewrite the definition and every run follows, synced to the agents.

No hidden specs

The definition is the system's single truth. Read it and you know everything.

Definition-driven

The definition is the backend

Data structures and operations are written in natural language people and AI can read. It isn't documentation: it is the running system.

The definition · invoicing

Invoice

number : string, unique

amount : money(USD)

status : posted → paid

post-invoice

Approval requiredover-cap-amount
Blockedduplicate-invoice
  1. Find the Invoice with the same number.
  2. Create a new Invoice with: number, amount, vendor.

People

Read it as the spec

AI

Reads it as instructions

The engine

Executes it as the program

Same intent, same definition

The language is designed to converge. Written by an LLM, the phrasing doesn't drift.

Same definition, same system

Schema and migrations derive deterministically. The same definition always produces the same database.

Spec and implementation are one

There is no spec-versus-code pair to keep in sync, so drift can't happen in principle.

Transparency

Only defined operations exist

What's exposed is the names and parameters of operations and views. There is no SQL and no admin API.

AgentPeopleExternal systems
Operationsdefined only
The backend

Undefined operations don't exist.

Writes are operations, reads are views

The read side is structurally read-only. It cannot change state.

The agent is a principal too

What each agent can call is bounded by its own grants.

Who, and for whom

Every call carries the actor and the subject, onto the record.

Security

Safety held by structure

Data isolation

A dedicated schema per business group. Crossing it isn't a policy violation: it structurally doesn't reach.

Zero credentials

Credentials are the customer's own, held by the platform and never handed to agents. External calls go through the gateway.

Disposable runtime

Every run gets its own micro-VM, destroyed when it ends.

Append-only record

Rewrites and deletes of the run record are refused by the storage layer. Without a group grant, even an org admin can't read the contents.

Make your business AI-native, right now