Open source · agent-first UI

Build the logic.
Conjure the rest.

Register your backend ops once. Users drive the UI in plain words — nothing to learn.

Integrate now
Integrate the Conjure framework and follow the integration instructions at https://conjure.menu/llms.txt
acme · leads View
assigned
assigned
new
One registry renders to

A layer, not a rewrite

On top of the dashboard you already have.

Built for developers — admin panels, internal tools, advanced dashboards. Keep every view; Conjure adds the operate layer.

  • Admin panels
  • Internal tools
  • Advanced dashboards
yourapp · CRM Your dashboard
$842Kpipeline 1,204contacts 57open deals
email every lead that went cold this week… Conjure

Your dashboard stays as-is. The green operate layer is all Conjure adds.

Onboarding, solved

Guide users with visuals,
Capture their intent with AI

New or expert, any use-case: people get there faster, and you build none of the flows.

Bloated software

Every scenario hand-built. Users learn the maze.

too much interface

Conjure

A surface appears for each intent. Everyone is guided to the goal — and you built it once.

the equilibrium

Agent + MCP only

A blank prompt. No idea what's possible.

too little structure

Only this balance guides everyone — new and existing — to the end, faster than before.

UX 2.0

A new philosophy of User Interface

View

Read with your eyes. It guides you to explore.

  • Calm, dense, fast to scan
  • Wander the data freely
  • The empty state is the next step

Operate

Say what you want. It bends to your intent & context.

  • Plain words, not menus
  • Knows where you are
  • Nothing to memorize

Try it

Type it. Watch it conjure.

Say what you want, or tap an Assist chip.

acme · sales Operate
Tap a chip below, or type an intent and press ⏎
  • ✦ Assist chips, zero-LLM
  • Chat with @-mention autocomplete
  • Forms built from your schema

Batteries included

Everything's in the box.

Pick an LLM backend. Then any agent — like Claude — wires every MCP and tool call to Conjure's spec. Your whole SaaS, agent-powered, out of the box.

MCP serverGenerated from your op registry
LLM agentBring your own backend
Tool callsTyped, gated, reversible
Assistant drawerChat, @-mentions, autocomplete
Assist menuContextual ✦ actions, zero-LLM
Dynamic forms & cardsBuilt from your schema
1
Choose an LLM backendAny provider — your keys.
2
Point your agent at Conjure“follow conjure.menu/llms.txt”
Your SaaS is wiredEvery MCP & tool call, to spec.

Demand analytics

The holy grail: what users actually want.

Every action is a stated intent — the ultimate read on demand. See the features users ask for that don't exist yet, and exactly where they get stuck.

modelstat · demand Analytics
2,481intents / wk 37%have no feature ↑ 24%vs last month
Top intents — no feature yet
  • “merge duplicate contacts”312
  • “bulk-email a segment”248
  • “forecast this quarter”190
  • “auto-assign by territory”154
  • “export deals to Sheets”121
Where users get stuck
  • “can't filter won deals”88
  • “import keeps failing”64
  • “no undo on bulk edit”41
  • “duplicate contacts everywhere”33

Built from intent, redacted on-device. ModelStat SDK · privacy-safe · optional add-on.

Get started

Nothing to build. Nothing to learn.

Register your ops once. The UI writes itself — users act in plain words, with nothing to learn.

ops/leads.ts
defineOp({
  name: "leads.assign",
  method: "POST",
  path: "/v1/leads/assign",
  summary: "Assign a lead to a rep",
  effect: "write",
  inputSchema: { required: ["rep_id"] },
})
Integrate now
Integrate the Conjure framework and follow the integration instructions at https://conjure.menu/llms.txt

prefer manual? npm i @modelstat/conjure · GitHub · MIT