disidea
Multi-model decision workspace

Stop asking one model
and hoping.

Put a decision to four frontier models in a single thread. They read each other, disagree in the open, and leave you with the argument — not one confident answer you have no way to check.

See pricing
No credit card. The free plan is enough to watch them disagree.

The panel

Four models, each with a standing job. Swap any of them for another the gateway tracks.

O
GPTStructures the decisiongpt latest
✦
GeminiSynthesizes contextgemini pro latest
K
KimiTests practical alternativeskimi latest
A
ClaudeChallenges assumptionsclaude opus latest

Three ways to run the panel

Conductor

One model reads the request first and decides how much panel it deserves — often none.

Simple questions get answered on the spot for one credit. Real ones get a written plan, and each step goes to the model best suited to it.

Sequential

Each model reads every answer above it before writing its own.

Earlier models inspect and review; only the final model can edit files, after reading the whole chain.

Parallel

Every model answers independently, with no knowledge of the others.

For code, each model gets the same baseline and its own private candidate. Compare the diffs, then choose one result without merging.

It does the work, not just the talking

A project the models can edit

They list, search, read and write real files — up to 60 per thread, 1.9 MB each. Every change is recorded per turn, with a diff and a one-click restore.

Previews that cannot touch you

Generated code runs in a sandbox on a separate origin with no bindings, no cookies and no access to your session. Markdown and HTML render as documents; React is bundled in your browser.

A brief that keeps up

A separately configured utility model reads the thread: the current verdict, its confidence, what the panel agrees on, what it still does not, and what to do next.

Stop means stop

Ending a run reaches the provider, not just your screen — tokens stop being generated and billed. Whatever had already arrived is kept.

Your thread, on your disk

Export any thread as Markdown, with every answer, every tag and every file a model touched. Upload a PDF and its text is read into the project for the panel to use.

You pay what it cost

A credit is $0.01 of model time, billed from the figure the provider actually reports — not a flat charge per message. A quick question is about one; a four-model panel about four. A run stops when the balance does.

Pricing

A credit is $0.01 of model time, and you spend what a message actually costs. A quick question is about one credit; a four-model panel is about four; a coding turn that reads your files is about twelve. No plan is unlimited, because nothing here is.

Free

Free

10 credits a month

Relative
Base usage
Up to
5 messages
Or about
2 panel runs

Enough to watch four models disagree, which is the part worth paying for.

  • 10 credits a month · about 2 panel discussions
  • Up to 5 messages a month
  • Conductor and Sequential modes
  • One Scribe brief a month
Best to try

Basic

$9/mo

250 credits a month

Relative
25× usage
Up to
250 messages
Or about
62 panel runs

For putting real decisions through the panel a few times a week.

  • All three orchestration modes
  • The full four-model panel
  • File tools and the preview sandbox
  • Unlimited Scribe briefs

Max

$99/mo

3,000 credits a month

Relative
300× usage
Up to
3,000 messages
Or about
750 panel runs

For building with it all day. Past this, talk to us rather than guess.

  • Everything in Pro
  • Covers a full-time building workload
  • Fast customer support — replies within 24 hours
  • Highest tool-round limits
Prefer not to subscribe?

Pay as you go

Buy credits once. There is no renewal, and purchased credits never expire.

100 credits

$5 one time

A small one-off top-up.
1,000 credits

$39 one time

For heavier work without a subscription.

Cancelling keeps your threads: conversations stay readable and exportable for as long as your account exists, and project files stay for at least six months after a plan ends. Message counts are a ceiling, not a forecast: they assume every message is the cheapest sort, so a month spent on four-model panels lands nearer the second line than the first. What is billed is the real cost the provider reports for your calls, to the hundredth of a credit — never a flat charge per message. Plans are monthly; cancelling keeps the plan running until the period you already paid for ends.

Read before you sign up

Questions people ask first

Is it safe to let a model write and run code?

Generated code never runs in the application. Every preview is bundled in your own browser and executed in an isolated container on a separate domain — no cookies, no session, no access to your account or your data, and no network reach beyond the package registry it needs to load libraries. It cannot call out to anywhere else, and it cannot see anything of yours that you have not put in that project. Four independent boundaries sit between model-written code and your account; if any one of them failed, the other three would still hold.

How the sandbox is isolated

What happens to my work if I cancel?

Your threads stay. Every conversation, every answer and every model's reasoning remains readable and exportable for as long as your account exists — cancelling a plan does not take away what you have already produced. Project files stay for at least six months after a subscription ends, so there is no rush to move anything. You can export a thread as Markdown, or download a project, at any time and without asking us.

Still deciding?

Ask us anything about the modes, the models, or which plan fits what you do. Pro and Max are answered within 24 hours.

support@disidea.dev

Bring it a decision you have been putting off.