SHELLEDAI
Try Shelled

ONE PROMPT · ONE OMNI AGENT · ONE REVIEWED PULL REQUEST

Software that builds itself.

Shelled AI turns one prompt into working software. Tell the Omni agent what to build; it plans, codes, tests and reviews with a team of specialist agents, then hands back a live preview and a reviewed pull request.

Try Shelled How a build runs

01 / WHAT IS SHELLED

An AI software team you talk to through one agent.

Shelled AI is a platform for building software with coordinated AI agents. Instead of prompting a single model and stitching the answers together yourself, you describe the outcome you want and Shelled runs the whole job: it plans the work, writes the code, runs the tests, reviews the diff and prepares the release.

You only ever talk to one agent, Omni. Omni breaks the request into tickets, routes each ticket to the right specialist and keeps every agent working from the same project context, so decisions made during planning are still visible when the code is reviewed. The result is not a chat transcript. It is a working preview and a pull request your team can read, test and merge.

Shelled is built for the distance between an idea and production. The workspace, the agents, the harness that lets them act, the models they reason with and the infrastructure they run on are designed as one connected system rather than a set of separate tools.

PRODUCT
Shelled Cloud, at app.shelled.ai
INTERFACE
One chat with the Omni agent
UNDER THE HOOD
Architect, Coder, Tester, Reviewer, Deployer
OUTPUT
Live preview + reviewed pull request
RUNS ON
Cloud, local and private environments
BASED IN
San Francisco

02 / HOW A BUILD RUNS

Seven modes, from the first sentence to a merged change.

Every Shelled run moves through the same modes you see along the bottom of the app: Plan, Reason, Build, Test, Draft, Draw and Preview. Omni switches between them as the work demands, and returns to earlier modes when a test fails or a review asks for changes. Each mode leaves evidence behind, so you can see exactly how the result was produced.

  1. 01 · PLAN

    Plan

    Shelled reads the request and the repository, then writes down what done means before touching code. The plan names the files, the interfaces and the open questions, so the rest of the run has a target to check against.

  2. 02 · REASON

    Reason

    Tradeoffs get worked through explicitly: data model, dependencies, edge cases, failure modes. Reasoning stays attached to the project, so later steps can see why a decision was made instead of rediscovering it.

  3. 03 · BUILD

    Build

    Agents implement the change across the interface, application logic and data layer. Independent tasks run in parallel as tickets, and each ticket carries its own context, diff and status.

  4. 04 · TEST

    Test

    Checks run against the change. Failures are read, investigated and fed back into the next attempt rather than reported and abandoned. Test evidence is kept with the work for review.

  5. 05 · DRAFT

    Draft

    Copy, documentation, release notes and pull request descriptions are written from the actual diff and the plan, not from a generic template.

  6. 06 · DRAW

    Draw

    Interfaces, diagrams and visual assets are produced inside the same workflow, so design and implementation stay in sync instead of drifting between tools.

  7. 07 · PREVIEW

    Preview

    The result is rendered and inspected before it is handed back. You see a working preview and a reviewed pull request, ready for your team's merge decision.

03 / COORDINATED INTELLIGENCE

One Omni agent. A team of specialists behind it.

A single model asked to do everything tends to lose track of the plan, skip the tests or review its own work too kindly. Shelled splits the job into roles. Each specialist has a narrow responsibility, its own tools and a clear handoff to the next, and all of them write into the same pull request.

  • Architect

    Reads the repository and turns the goal into a plan: scope, structure, interfaces and the order of work.

  • Coder

    Implements across the stack and coordinates independent tasks so they land in the same branch without conflict.

  • Tester

    Runs checks, investigates failures and feeds concrete results back to the Coder until the change holds up.

  • Reviewer

    Examines the diff against the goal, the project's conventions and the test evidence before anything is proposed.

  • Deployer

    Prepares the reviewed pull request and the release handoff, with a clear summary of what changed and why.

Because the specialists share one context, a ticket can run in parallel with others without drifting from the plan. When the Tester finds a failure, it goes straight back to the Coder with the failing output attached. When the Reviewer asks for a change, the reason is recorded next to the diff.

04 / THE HARNESS

Give intelligence a way to act, and a way to be checked.

A model on its own produces text. The harness is what turns that reasoning into useful work: it gives agents tools, carries context between steps, limits what they are allowed to do and makes them prove the result. The path from the plan to the pull request stays visible the whole way.

Tools
Repository access, a shell, package managers, browsers and previews, so agents can act instead of only suggesting.
Context
The plan, the codebase, prior decisions and test results travel with every ticket.
Permissions
Agents work inside defined boundaries. Nothing merges without your team's decision.
Checks
Tests, type checks and builds are run, read and acted on as part of the loop.
Review
Every change is examined against the goal and the project's conventions before it is proposed.

Read the full harness overview

05 / MODELS & INFRASTRUCTURE

The right model for the work, wherever the work lives.

Shelled connects models for reasoning, coding and vision to a shared project context. Reasoning models explore the problem and the tradeoffs. Coding models implement with the repository and tools in view. Vision models bring screenshots, designs and interface context into the same workflow. The surrounding system supplies the instructions and evidence that make each answer useful.

The project and its people stay at the center while the infrastructure around them changes. Teams can start in the cloud and move sensitive work closer to home without rebuilding how they work.

  • Local

    Work close to your local repository, tools and chosen models.

  • Cloud

    Shelled Cloud connects cloud intelligence to the project, its tools and its context.

  • Private

    Shape a private or on-prem deployment around your team's security and data requirements.

06 / WHAT YOU CAN BUILD

From a SaaS product to a fifty-agent research run.

New projects in Shelled Cloud start from one of four templates, or from an existing repository you import. The template sets the specialists, tools and checks that fit the kind of work.

  • General project

    Full-stack products in TypeScript and React: authentication, billing, APIs, dashboards and the glue between them. Start from a sentence or import an existing repository.

  • Immersive site

    Marketing sites, interactive stories and WebGL experiences, where design, motion and performance all need to be right at once.

  • Math + Lean research

    Formal mathematics with the Lean proof assistant. Agents explore conjectures, write proofs and let the checker decide what is actually true.

  • Erdős–Straus · 50 researchers

    A research-scale run: fifty coordinated researcher agents working on the Erdős–Straus conjecture, sharing partial results through one project.

07 / QUESTIONS

Frequently asked questions about Shelled AI.

What is Shelled AI?

Shelled AI is a software-building platform from San Francisco. You describe what you want in one prompt, and a single Omni agent coordinates a team of specialist agents to plan, build, test and review the work, returning a working preview and a reviewed pull request.

What is Shelled Cloud?

Shelled Cloud is the web app at app.shelled.ai/cloud. It is where you create or import projects, chat with the Omni agent, watch tickets execute in parallel and review the results.

What is the Omni agent?

Omni is the single agent you talk to. Behind it, Omni routes work to specialists (Architect, Coder, Tester, Reviewer and Deployer) and keeps their work in one shared project context, so you never have to manage the agents yourself.

What does a Shelled run produce?

A working preview of the result, the code changes as a reviewed pull request, the test evidence and a written summary of what changed. Your team stays in control of the merge.

Can I use Shelled on an existing codebase?

Yes. Use Import existing project in Shelled Cloud to bring in a repository. Agents read the codebase and follow its conventions instead of starting from a blank template.

Which models does Shelled use?

Shelled connects models for reasoning, coding and vision to a shared project context. The harness around the model carries the tools, instructions and evidence that make its output useful, so the right model can be used for each part of the work.

Where does Shelled run?

Shelled runs in the cloud by default and is designed to connect local tools and models, cloud models and private or on-prem environments, so teams can keep code and data where their requirements say it belongs.

Who is Shelled for?

Independent builders who want to ship a product from an idea, engineering teams who want parallel, reviewed execution on real repositories, researchers working in formal mathematics, and companies that need agent work inside their own infrastructure.

Your next idea. Try Shelled.

Open Shelled Cloud, type what you want built and watch the tickets run. Bring an existing repository or start from a sentence.

Open Shelled Cloud