Quadrazene
Quadrazene™

Authoring workbench

Compose.
Describe it. Commit it.

One front door for building everything. Describe what you need in plain English. An agent decomposes it into a Skill, a Formula, a View, or a one-shot Insight. A live preview shows the working draft. Click Commit to promote it to a real, governed artifact.

quadrazene.app/compose
quadrazene v0.2.16
Quadrazene
ReactorComposeRecipesInboxGovernanceRecords
⌘K
Conversation
You
I want a saved query for pipeline coverage broken down by region and quarter, with a KPI tile and a waterfall.
Compose agent
Insights
Drafting a Skill bound to nl-query-to-insight with two parameters and a View.
draft_skill()draft_view()revise_draft()
You
Use the cheaper model for the summary step, and rename it “Pipeline Coverage.”
Type to refine the draft… ⌘↩ to send
Live preview · draft
uncommitted
InsightsPipeline Coverage
Show pipeline coverage broken down by {{dimension}} for {{timeframe=this quarter}}.
Formula
nl-query-to-insight
View
kpi + waterfall
DiscardCommit skill

Two panes: the conversation (left) and the live draft preview (right). Commit is admin-gated.

One interface, every artifact

Stop hunting for the right editor.

Compose drafts every kind of Quadrazene artifact in one place, with a renderer that previews exactly what you'll get.

S

Skill

A parameterized entry point. Prompt template + Formula binding + optional View. Lands in the Reactor rail.

F

Formula

A sequence of atoms with sequential / parallel / branch / forEach steps. Versioned and replayable.

V

View

A layout spec for the result. KPI tiles, tables, waterfalls, action buttons. Previewed block-by-block.

O

One-shot Insight

Run a question right now without saving anything. Answer streams into the conversation.

How it works

An agent with tools, not a text box.

01
Describe
Say what you need in plain English.
02
Decompose
The agent calls draft_skill / draft_formula / draft_view.
03
Ground
Atom registry + your Formulas + tenant context injected per turn.
04
Refine
“Add an error handler.” “Use the cheaper model.” Multi-turn.
05
Commit
Promote the draft to a real, versioned artifact.

Grounded + governed

It can't invent atoms or publish behind your back.

Grounded in your catalog

The system prompt is rebuilt each turn with the live atom registry, your existing Formulas, and your tenant context, so drafts reference real building blocks, not hallucinated ones.

Admin-gated commit

Anyone can draft, only admins can commit. Non-admins hand off a ready-to-publish draft.

Safety rails

Hard caps on agent loop iterations, output tokens, and wall-clock time keep every session bounded.

Session persistence

The full conversation and draft survive navigation. Come back and pick up exactly where you left off.

Per-step model choice

Pin a specific LLM profile to an individual step right from the conversation.

Preview before publish

Per-artifact renderers (SkillCard, Formula tree, View preview, Insight card) show the real thing before commit.

Prefer to drag boxes? See the visual Designer →

Beyond the first draft

What Compose builds on now.

Tenant context

Industry, regulatory tags, primary systems and risk posture are configured once for the tenant and shape how Compose decomposes a request.

Agent-team steps

A Formula can hand its plan to a supervisor that dispatches specialist Formulas in parallel and synthesizes the result.

Immutable once published

Published versions are locked. Change means a new version, so what ran yesterday is still what it says it was.

Ship it as a bundle

What you author can be exported as a signed Reagent bundle and installed into another tenant or deployment.

Traced end to end

Every LLM call Compose makes is measured and nested in the trace, with cost, latency and scores.

A shared model picker

One picker chooses models everywhere, from a curated catalog with a live refresh against the provider.

Describe your first workflow.

We'll open Compose against a sample of your data and build a real Skill with you, live.

Request a demo