Free desktop app · Mac and Windows

Run an AI development team
on your own projects.

Give Cohandry an objective, from one feature to a complete game, and a team of agents works on it autonomously. A director plans, builders write and test the code, and a reviewer checks every change. You step in only when a decision needs you.

Projects

Task Garden Active

Active. New work starts when its requirements are met.

Next step

Tell your team what to build

Give the director one bounded objective. It plans, delegates and gathers reviews; you make the final call.

Interactive product walkthrough · Simulated team, no model calls.

Simulated Unity Game Studio project · Pack coming soon · No model calls.

The app is free and you bring your own model, so the only usage bill comes from your provider.

Development preview · Apple Silicon Mac · Windows x64 · Intel Macs are not supported.

Why a team

When one chat writes the code and grades it too, nothing catches its mistakes.

What it’s like now

  • A wrong assumption in the first message survives all the way to the final answer, because the same model that made it is the one checking the result.
  • Every new session starts cold, so you paste the same files, conventions and decisions back in before any real work begins.
  • In Unity you become the glue between the chat and the editor: you copy the C#, wire the prefab, press Play and paste the console error back by hand.
  • When the answer ends with “it should work now,” that sentence is the only proof you get.

With Cohandry

Cohandry gives each job to a different agent and keeps the evidence for every change.

Separate roles
A director plans the work, builders make the changes and a reviewer checks them. Because each role can run on a different model, the reviewer is never grading its own homework.
Evidence on every change
Every submission arrives with its diff, its test output and the reviewer’s findings for that exact revision, so you approve what you can actually see.
Your call where it matters
Commands, merges and pull requests wait for your approval, while the work in between runs on your own computer without you watching every step.
Packs for specific stacks
A pack swaps in a team and automated checks built for one kind of project, starting with Unity, so reviews are judged against compile and test results instead of a description.

From a folder to reviewed work in three steps.

You set the goal and make the final call, and the team handles everything in between, from the first plan to the reviewed diff.

  1. 1

    Pick a team

    Start from a free template for software, research or creative work, or from a pack built for your stack, and adjust the roles to fit.

  2. 2

    Connect your project

    Point it at a project folder, optionally a GitHub repository, and a model from OpenAI, Anthropic, xAI or OpenRouter.

  3. 3

    Hand off one objective

    Describe what done looks like and the team plans, builds, tests and reviews it. Once you approve, it can open a draft pull request on GitHub.

See every stage in detail

1Set up

Point a team at your project folder.

Choose a starting team, a model and the folder or repository it works in. Setup shows each item exactly as it is, so you can tell a saved setting from one that is actually connected and tested.

2Direct

Give the director one bounded objective.

Say what done looks like, and the director turns it into a plan, splits it into assignments and decides who reviews what. Your brief and project documents travel with every assignment automatically.

3Team works

Follow the handoffs instead of scrolling a chat log.

Each agent shows what it is working on and what it is waiting for, and you can open any task in VS Code or Cursor or step into its terminal. Every command waits for your approval before it runs.

4Review

A reviewer checks the evidence before you do.

Each submission arrives with its diff and test output. The reviewer inspects that evidence, records findings against the exact revision and sends focused fixes back to the agent that wrote the code.

5Your call

You decide what ships.

With GitHub delivery configured, approved work goes to a draft pull request where you can check CI and reviews before merging. If you request changes instead, the feedback goes straight back to the team and the work so far is kept.

Packs · Coming soon

Describe your game. Get playable, reviewed Unity milestones.

The free app gives you a general-purpose team. Packs replace it with a team built for a specific kind of project or tool, such as a Unity or Godot game, with its own workflow, tool integration and automated checks. Packs are made by Cohandry and downloaded separately from the app.

Unity 6 · C#Coming soon

Unity Game Studio

You describe the game in a paragraph and the director turns that conversation into a design document you approve. The team then builds it milestone by milestone until each one is ready for you to play.

How packs will work

Nothing is on sale yet; the price and release date will be announced at launch. Each pack will be a one-time purchase that includes its future updates.

Agents

  • Game directorRuns the design conversation with you, writes the game design document and roadmap, and splits milestones into small, testable tasks.
  • Unity implementerBuilds each task in C#, scenes and prefabs, runs the checks and submits the evidence.
  • Code reviewerChecks every revision against its acceptance criteria and the check results, then approves it or sends back precise fixes.

Workflow

  1. Design — a conversation with the director
  2. Approve GDD — you lock the game design document
  3. Prepare Unity — connect the Unity project and Git
  4. Build milestones — each task built, checked and reviewed
  5. Playtest — you play each milestone before the next starts

Automated checks

  • Compiles in Unity
  • Console checked for errors
  • EditMode and PlayMode tests
  • Screenshots captured for review

These checks run inside Unity after every task. By default the reviewer approves a task only once they pass, and you approve each finished milestone yourself.

Requirements

  • Unity 6
  • The Unity CLI
  • The Unity Pipeline package
  • A model connection of your choice

More packs and free teams

Godot · GDScriptPlanned

Godot Game Studio

The same design-to-build workflow, adapted for Godot projects and GDScript.

Agents to be announced

Any stackFree · included

Software delivery

Takes a feature from your description to a reviewed pull request.

Technical lead · Implementation agent · Code and test reviewer

Docs · sourcesFree · included

Research and reports

Takes a question to a report where every claim cites its source.

Research lead · Researcher · Evidence reviewer

See all templates →

Pick the kind of work you want done and see how the team gets there.

Most people start with a coding task, then hand the same team a bug, an internal tool or a technical decision once they have seen how the handoffs work.

WEB APPS & INTERNAL TOOLS

Take an app from idea to working feature.

Whether it is a dashboard, a small web app or an internal tool, the lead turns your request into a plan, an implementation agent writes the code and tests, and a reviewer checks the result before you see it.

Your next objective could start like this

“Add saved views to this task dashboard. Include keyboard navigation, tests and a reviewable pull request.”

All you need to start is a project folder, your usual development tools and a connected model.

Works with your editor, repo and tests.

Cohandry works alongside your editor and repository rather than replacing them. The common connections are built in, and you can give each agent more tools through MCP servers you configure yourself.

  • GitHubTask branches, draft pull requests, review feedback, CI checks and merge decisions.
    Built in
  • VS Code & CursorOpen the task’s workspace in your editor and pick up where your agents left off.
    Built in
  • Terminal & test toolsAgents request project commands and you follow the output, or step in with the owner shell.
    Built in
  • Configured MCP serversConnect additional applications through tools you explicitly configure and allow.
    Optional
How should your team think?
DemoSimulated work. No model, no cost.Provider APIUse your chosen provider and model.Optional CLISupported CLI with its own sign-in.
OpenAIAnthropicxAIOpenRouter

Set a project default, or give each agent its own model.

GitHub, editor shortcuts and the owner shell are built into the workflow. MCP servers and external development tools are installed and configured on your execution host; available actions depend on the connected tool.

Enterprise edition · Preview

Run Cohandry on your company's AWS account.

The Enterprise edition is the same desktop app, but its AI access, safety rules and audit trail come from your organization's AWS account rather than from personal API keys, while projects and code stay on each person's computer. It is in preview and not part of the public download yet.

Company sign-in

People sign in through AWS IAM Identity Center, and the app holds only short-lived credentials that its agents never see.

Models on Amazon Bedrock

Agents use the models and region your administrators approve, including OpenAI's gpt‑oss models, billed to your AWS account.

AWS-enforced guardrails

IAM refuses any model call without your company's Bedrock Guardrail, and tool results are screened before a model reads them.

CloudTrail audit trails

Every model call is recorded under the name of the person who made it, next to the rest of your AWS activity.

Questions and specs

Still wondering about something? .

At a glance
PlatformsApple Silicon Mac, Windows x64
Where work runsOn your computer, in your project folder
ModelsOpenAI, Anthropic, xAI and OpenRouter APIs, or a supported CLI with its own sign-in
Your dataProject records stay local. Only the context an assignment needs goes to your model provider.
ApprovalsCommands, merges and pull requests wait for you
PriceFree app. Paid packs coming; provider usage billed by your provider.
What can I build with Cohandry?

Anything you would hand to a small engineering team: new apps and features, bug fixes, refactors and technical research. You give the director an objective, the agents move it through implementation, tests and review, and with GitHub connected the result lands in a draft pull request.

Can I use it with my existing projects?

Yes. Connect your repository, select a local project folder and configure the agents and tools for your stack. Open task workspaces in VS Code or Cursor, inspect changes and use the owner shell to work alongside your agents.

How do external tools connect?

Configure an execution host with an installed MCP server and the tool names your agents can use. Cohandry can then route assignments through those tools. Available actions depend on your configured server. Each server provides its own actions and setup; the built-in file workflow is available for ordinary coding tasks.

Is Cohandry free?

The base app is free to use, and no card is required. Bring your supported model subscription or API connection; provider usage and external services have their own pricing.

What is a pack, and when can I get one?

A pack is an add-on made by Cohandry for one kind of project: a ready team, its workflow and checks built for that work. Each pack will be a one-time purchase that includes future updates to that pack, downloaded and installed separately from the free app. Packs are not available yet, so nothing can be bought or installed today. Unity Game Studio is first; its price and release date will be announced on the Packs page. Your projects and the free app never depend on a pack.

Which models can I use?

Connect a model-provider account, or use provider APIs from OpenAI, Anthropic, xAI and OpenRouter. Set a default for the project, or choose a different connection for each specialist. Access and tool support depend on your selected account and model.

What do I need to get started?

Downloads are available for Windows x64 and Apple Silicon Mac and include the application runtime. Connect your model account, select your project folder and configure the tools you want agents to use. Read the Task Garden tutorial for a complete coding project, from the first assignment to GitHub delivery. The current Mac build is ad-hoc signed and not notarized; see Preview notes for build details.

Where does my project data live?

Your project records are stored in the local coordinator’s data directory. The context needed for an assignment goes to your chosen model provider. GitHub and other enabled integrations receive the information needed for their part of the workflow.

Hand off your next feature.

Start with one feature, bug or research question, and see the plan, the diff and the review before you approve anything. The app is free to download.

Free base app · Email verification required · No card required

Apple Silicon Mac · Windows x64 · Intel Macs are not supported. · Development preview
THE COHANDRY PREVIEW

Your next build
starts here.

Verify your email to download the free desktop preview. Links expire after five minutes and can be refreshed. No password. No card.

We'll use your email to verify download access.

YOUR NEXT BUILD

What are you
planning to build?

An optional answer helps us understand what matters to you. Both questions are optional and do not affect download access.

A sentence or two is plenty. Please leave out private code or confidential project details.

WE’RE LISTENING

Make the next
version better.

Tell us what happened, what you’d like to see, or where you got stuck.

Used only to follow up on this report. This does not subscribe you to product updates.

Up to 2 files, 3 MB each. PNG, JPG, WebP or plain text. Review files for private information before attaching; reports stay private.

By submitting, you’re sending this report to the Cohandry team.

COHANDRY