PLUGGIE · PRIVATE BETA

Your agent defines the backend.
We run everything else.

Pluggie builds platform tools for teams who ship with AI agents. AgentX, our flagship, turns one MCP conversation into a branded client admin and a production delivery API — no per-project backend code, ever.

Become a beta tester Read the docs →

0MCP tools0endpoint families0smoke tests green

mcp · agentxconnected

/ 01

One conversation. A whole backend.

step 01

Agent defines the schema

One define_collection call. 8 field primitives, constraints, computed fields, workflows — a closed, self-describing vocabulary agents can't misuse.

define_collection({

name: "posts",

fields: { title: "text!",

slug: "slugify(title)" }

})

step 02

A branded admin appears

Your client gets a polished, Clerk-gated admin in their brand — entry forms, media, trash, version history. A hand-off artifact, not an internal tool.

Meridian Galleryadmin

Autumn exhibitionpublished

Open call 2026draft

Artist residencydraft

step 03

The site consumes the API

Per-field public reads, search, relations, i18n, ETags — plus a typed TS client generated from the live schema, compile-verified under strict.

GET /v1/posts?locale=fr

200 etag: "a9f2…"

{ items: [{ title:

"Exposition d'automne" }] }

/ 02

Everything a backend needs. Declaratively.

explore AgentX →

001Data modeling8 primitives, constraints, computed fields, schema diffs with plan + confirm.

002Delivery APIPer-field public reads, query power, keyword search, i18n, strong ETags.

003AuthorizationFail-closed presets, BYO Clerk, owner rows, org scoping — no expressions.

004AutomationEvents, delayed actions, schedules, declarative state machines.

005PaymentsStripe checkout from entry ids, order lifecycle, declarative fulfillment.

006ComputeBefore-write hooks on your infra. AgentX never hosts your code.

007RealtimeChange feed + SSE with then-AND-now privacy gating.

008TrustTrash + restore, version history, audit log, plan + confirm everywhere.

/ 03

Built for agents. Not adapted for them.

Every surface is machine-legible by design. An agent can repair its own mistakes from the error alone — no human in the loop.

Errors that fix themselves

E_VALIDATION

{ field: "price",

constraint: "min", limit: 0,

hint: "price must be ≥ 0;

received -4" }

Structured ConstraintIssue[] on every failure, from an append-only E_* registry.

Destructive = plan + confirm

plan:

− drop field "subtitle"

(212 rows affected)

+ rename "body" → "content"

confirm to apply

Schema diffs, purges, delocalization — nothing destructive runs without a plan.

Safe to retry, safe to race

update_entry_if({

if: { stock: 3 },

set: { stock: 2 },

idempotencyKey: "ord_91" })

CAS writes, multi-op transactions with dry-run, idempotency receipts on replays.

/ 04

The product family

The MCP-native backend platform. An agent defines the data model; a branded admin and a delivery API appear. Bring your own Clerk, Resend and Stripe.

Hostile AgentCOMING SOON

An autonomous security-testing agent for systems you own. It tries to break your rules before someone else does — and reports what it finds.

join the waitlist →

PRIVATE BETA

Give your agent a backend it can actually drive.

We're onboarding a small group of agencies and AI builders. Deployed, tested, and running production client sites today.

Become a beta tester