Who is Foo?

SAY WANT YOUWANT, FOO!

...we’ll make sure your Agent understands it.

SayWhat compiles rough instructions into execution-ready assignments — and learns your preferred rules, constraints, and working style over time.

Try PreflightSee how it works

Your agent heard you. It just shouldn’t start yet.

Preflight

You typed this...

make this dashboard less crappy on mobile and use the card thing from settings but don't mess up auth

SayWhat.foo returns this...

✈️ SayWhat.foo PreFlight

What SayWhat added
inspect the dashboard implementation; reuse existing Settings card components where appropriate
Compiled assignment
Improve dashboard mobile responsiveness while preserving existing authentication behavior.

One preflight layer. Any coding agent.

  • Cursor
  • Claude Code
  • Codex
  • Windsurf
  • VS Code
  • Copilot
  • Cline
  • Zed
  • JetBrains
  • Replit
  • TRAE

How Preflight works

SayWhat helps you improve intent before AI acts on it.

Coding agents are powerful execution engines. They still make assumptions when instructions are incomplete. Preflight clarifies intent, constraints, context, and verification before execution — and your preferences can carry into future runs.

You

make this dashboard less crappy on mobile and use the card thing from settings but don't mess up auth

SayWhat Preflight

COMPILING INTENT…

  • OBJECTIVE
  • DISCOVERY
  • CONSTRAINTS
  • IMPLEMENTATION
  • VERIFY
Agent-ready specification

ObjectiveImprove dashboard mobile responsiveness while preserving authentication.

Discover the Settings cards. Reuse them. Don't touch auth. Verify the authenticated dashboard on a phone-width viewport.

Foo at a laptop turning a messy prompt into clear AI instructions

Mascot

FOO TALKS TOO MUCH.
THAT'S THE POINT.

Foo is the big mouth on purpose — the incoherent human half of every agent chat. SayWhat is the compiler mouthpiece he built so the yap becomes an execution-ready assignment before the agent starts.

Who is Foo?

WHO THIS IS FOR.

You talk like a human. Your agent gets a spec.

  • Founders

    Ship the thing without writing a spec novel.

    Editorial ink portrait of a founder

  • Designers

    Describe the feeling. Get a change the agent can execute.

    Editorial ink portrait of a creative director

  • PMs

    Capture the request once. Don’t babysit the wording.

    Editorial ink portrait of a product manager

  • Vibe coders

    You shouldn’t need a PhD in prompting to fix a button.

    Editorial ink portrait of a vibe coder

  • Juniors

    Ask for help like a human. Send it like an engineer.

    Editorial ink portrait of a software engineer

  • Seniors

    Skip writing miniature specifications all day.

    Editorial ink portrait of a startup operator

Prompt examples

MESSY IN.
EXECUTABLE OUT.

  • Prompt: make this dashboard less crappy on mobile and use the card thing from settings but don't mess up auth. Compiled: OBJECTIVE — Improve dashboard mobile responsiveness while preserving existing authentication behavior. (Dashboard · compiled)
  • Prompt: make checkout better. Compiled: Improve the existing checkout experience while preserving the current payment infrastructure. (Checkout · compiled)
  • Prompt: the login keeps doing weird stuff. Compiled: Debug the existing login flow. Do not rebuild auth. (Login bug · compiled)
  • Prompt: make this look more premium. Compiled: Polish the current UI so it feels more considered, using the existing brand, type, and components. (Visual polish · compiled)
  • Prompt: add subscriptions. Compiled: Add subscriptions using the existing payment provider and billing patterns if they exist. Do not stand up a second payments stack. (Subscriptions · compiled)
  • Prompt: deploy it but don't break anything. Compiled: Deploy through the existing pipeline. Do not invent a new hosting setup. (Careful deploy · compiled)
  • Prompt: the settings page is messy on mobile. Compiled: Fix settings layout on small screens. Keep the current IA and components. (Settings · compiled)
  • Prompt: rename UserCard to ProfileCard and update imports. Compiled: Rename UserCard to ProfileCard and update all imports. Don't change behavior. (Rename · compiled)
  • Prompt: add another pricing tier. Compiled: Add a third pricing tier using the existing checkout flow. Do not create a second payment path. (Pricing · compiled)
  • Prompt: the login form should not leak api keys to the client. Compiled: Keep secrets off the client. Audit the login form for leaked credentials. (Security · compiled)
  • Prompt: rename the button. Compiled: Rename the named control and update references. Do not change behavior. (Copy · compiled)
  • Prompt: dark mode is broken on the billing page. Compiled: Fix billing dark-mode tokens against the existing theme. Don't invent a new palette. (Dark mode · compiled)
  • Prompt: the empty state looks sad. Compiled: Write a useful empty state with one next action. Stay in the current design system. (Empty state · compiled)
  • Prompt: webhook retries are eating us. Compiled: Make webhook handling idempotent. Do not double-charge on retry. (Webhooks · compiled)
  • Prompt: search is slow. Compiled: Profile the search query and add the missing index. Don't replace the search stack. (Search · compiled)
  • Prompt: the nav wraps weird on tablet. Compiled: Fix tablet nav wrapping. Preserve desktop layout and existing routes. (Nav · compiled)
  • Prompt: can we undo that last migration. Compiled: Inspect the last migration and roll it back only if the rollback path is safe. (Migration · compiled)
  • Prompt: onboarding dumps people into a blank dashboard. Compiled: Give first-run a default project and a next step. Don't rebuild onboarding. (Onboarding · compiled)
  • Prompt: the date picker is a mess in safari. Compiled: Fix Safari date-picker overflow using existing form primitives. (Safari · compiled)
  • Prompt: emails go to spam. Compiled: Fix from-address and auth headers. Don't switch email vendors in this pass. (Email · compiled)
  • Prompt: i need a loading skeleton on that table. Compiled: Add a skeleton that matches the table layout. Keep existing loading states elsewhere. (Loading · compiled)
  • Prompt: flaky test in checkout. Compiled: Stabilize the checkout test. Quarantine is not a fix. (Tests · compiled)
  • Prompt: the modal doesn't trap focus. Compiled: Trap focus, Esc to close, restore focus on exit. Don't replace the dialog. (A11y · compiled)
  • Prompt: don't let agents touch .env. Compiled: Keep .env, keys, and credentials out of agent context. Filter before upload. (Secrets · compiled)

Adapters

ONE PREFLIGHT LAYER.
ANY CODING AGENT.

Keep your coding agent. SayWhat isn’t another IDE. It’s the preflight layer in front of the agent you already use. The skill is the doorway. The compiler does the work.

  • Cursor

    Beta

    Preflight in the Cursor Agent workflow; review panel as fallback

  • Codex

    Beta

    Preflight in the Codex CLI / desktop workflow; CLI compile as fallback

  • Claude Code

    Beta

    Preflight as a Claude Code Skill + MCP tool, with CLAUDE.md always-on guidance

  • Kimi

    Coming next

    Coming next. Generic compiler output until a native adapter ships.

  • Windsurf

    Coming next

    Coming next.

  • Antigravity

    Coming next

    Coming next.

  • OpenCode

    Coming next

    Coming next.

Cursor and Codex today. Claude Code, Kimi, and others next.

CLI

UNIVERSAL
FALLBACK.

Use the CLI where a native Preflight hook isn’t available. Compile, display, copy. It does not launch third-party agents yet.

$ saywhat "fix checkout on mobile"

$ saywhat --agent cursor "fix checkout"

$ sw

Objections, heard.

YEAH, PEOPLE
SAY THIS.

We’ve heard the objections. Most of them treat the adapter like the product. It isn’t.

  • “I can write this myself.”

    SayWhat says · Fair.

    You can. You can also rebuild your own formatter, lint rules, Git hooks, and shell aliases. SayWhat makes better task definition the default instead of another thing you have to remember.

  • “It’s just a skill file.”

    SayWhat says · Yep.

    The integration is tiny. That’s the point. The skill file is only the adapter. Behind it is SayWhat’s pre-execution compiler: intent, constraints, context, playbooks, and rules. The adapter is tiny. The compiler is the product.

  • “The AI should already understand me.”

    SayWhat says · Fair question.

    Your agent is smart. That doesn’t mean it should start changing code while key constraints are still implied. SayWhat makes the assignment cleaner before execution begins.

  • “Why would I pay $11.98 for this?”

    SayWhat says · Fair.

    One misunderstood agent run can cost more than $11.98 in wasted time, tokens, reverts, unnecessary edits, and cleanup. SayWhat doesn’t need to save you hours. It just needs to prevent one bad run.

Try it once. Then decide.

Run Preflight on the next vague instruction. Messy in. Executable out.

Try Preflight

Pricing

The preflight layer
for AI work.

You’re not buying a plugin. You’re getting access to the SayWhat compiler. $11.98, one-time payment. No subscription.

SayWhat Pro

$11.98

Preflight for your coding agent “wingman”. Install once and forget it. Improvements happen on our side via the API. You get perfectly intented prompts.

  • Preflight compiler
  • Project-aware compilation
  • User rules and working preferences
  • Playbooks
  • Native adapters
  • Web compiler and CLI
Try Preflight first

After you pay

One paste. Then type normally.

  1. 01

    Check your email

    We send your activation code and a setup prompt made for your account.

  2. 02

    Copy the setup prompt

    Paste it directly into Cursor, Codex, Claude Code, Kimi, or another supported coding agent.

  3. 03

    Let your agent configure SayWhat

    It detects the environment, connects the adapter, activates your license, and verifies Preflight.

  4. 04

    Type normally

    SayWhat runs before execution and compiles your intent into an agent-ready assignment.

No manual MCP setup. No config hunting. No API-key wrangling.

Context + rules

CONTEXT WHEN
YOU OPT IN.

SayWhat gets smarter about how you work. Tell it once how you like agents to run — preferred verification, protected systems, recurring constraints — and future Preflights can carry those rules forward. Opt-in context (framework, branch, selected files) never uploads your tree. Secrets and `.env` are filtered.

FAQ.

It is the pre-execution compiler for AI coding agents. Rough human instructions become a structured, context-aware assignment — objective, discovery, constraints, implementation, and verification — before your agent starts work. Human intent in. Execution-ready assignment out.