The Pantion Dialogs

Pantion manakin mascot

Agent coding is fast.
Your intent is the bottleneck.

Pantion interviews you until there's nothing left to guess.

See it in action Get started

The five-second version

pantion-dialog

You "I want a booking app."

Pantion "What happens on double booking?"

You "First wins. Second gets alternatives." HARD

Pantion Locked. That decision won't be guessed later.

Spec tools help you write better specs. Pantion helps you find what's missing from them.

This isn't just a software problem

Different domains. Different questions. Same principle.

pantion-dialog

You "A 10-second intro for my YouTube channel."

Pantion "Should the camera move, or stay fixed?"

You "Slow zoom out, revealing the logo." HARD

Pantion Locked. No generator will improvise a camera move.

Dialog
Canon
Build

Three steps. One conversation.

01 — DIALOG

Ask until it's clear

You describe the goal. Pantion asks the next best question to remove ambiguity — one at a time. When new questions stop changing answers, your intent has converged.

02 — CANON

Approve and translate

The conversation becomes the source of truth: decisions, rejected alternatives, rationale — verbatim. You approve the Canon, then Pantion translates it into spec files your agent can build from.

03 — BUILD

Build, reflect, amend

Any coding agent builds from the Canon with far fewer assumptions. After building, reflect on what reality taught you. When requirements change, amend the Canon — append-only, never rewritten.

Your intent survives. Everything else is replaceable.
Model-agnostic. Tool-agnostic. Stack-agnostic.
A smarter model next year doesn't start over — it picks up where the last one left off.

It separates what's fixed from what's flexible

Non-negotiables and adjustable defaults — clearly labeled, so agents know what they can touch and what they can't.

It preserves rationale

"Why this, not that?" stays attached to the spec. When you revisit in six months, the reasoning is still there.

It reduces variance across runs

Same Canon, more predictable builds and rebuilds. Two agents reading the same Canon produce equivalent results.

Two modes: thorough or automatic

Normal mode resolves every ambiguity step by step. Automatic mode generates the full dialog from your description and marks conservative assumptions with ⚡. Pick the depth that fits the moment.

One command. Then talk.

$ npx @pantion/dialogs install --all

Describe what you want to build. Pantion takes over.

Get started →

Works with your tools

Claude Code Cursor Windsurf Gemini CLI Codex Anything that speaks MCP
Can't I just talk to Claude or ChatGPT?

You can. But an LLM stops asking when you seem satisfied. Pantion stops asking when ambiguity runs out. It systematically classifies every decision as fixed or flexible, and won't stop until there's nothing left to guess. The result isn't a chat — it's a spec that any agent can build from without assumptions.

Is it a spec generator?

No. Spec generators structure what you tell them. Pantion discovers what you haven't told them yet.

How is this different from Kiro or SpecKit?

Those tools help you structure requirements you already have. Pantion helps you discover requirements you don't have yet. Also: Pantion works beyond software.

Does Pantion contain an AI model?

No. Your agent brings intelligence. The Pantion Dialog brings structure.

What if I'm not technical?

Pantion adapts its tone to the user. A beginner gets plain language. A technical lead gets precision. The coverage stays the same — the conversation just fits you.

Will this slow me down?

It's a short pre-build step that saves rework later. Most teams feel payoff after the first "oh — good catch" question.

What happens when requirements change?

You amend the Canon. An amendment is appended — the original dialog stays intact. The change is explicit, traceable, and rebuildable.

Pantion manakin
Agent coding is here to stay.
The Pantion Dialog makes it reliable.
Read a real canon