Codex
Coding agent
BuilderSaved role
Implement the change.
A workspace for your AI agents
In developmentBuild with agents
that talk to each other.
Bring Codex, Claude Code, and other AI agents into one place. Give each a job, see how their work connects, and follow an idea as it takes shape.
Explore a guided concept, or download the desktop app for macOS, Windows, and Linux. Connected agents and browser control are in development.
Taking shape
Room for
the whole job.
An idea, a few notes, and the tools to build it. Keep them together on a canvas you can arrange as you go.
In the app today
Keep your notes, tools, and a browser preview in view. Save your workspace and return to the same layout.
What comes next
Give agents reusable instructions, let them talk to each other, and follow their checks in a browser. These features are in development.
Where we're headed
Give each agent a job.
See the work come together.
One agent builds. Another checks the result and asks for a fix. Try this guided example to see how their conversation could work.
Interactive concept. This is a demonstration of planned features.
Concept walkthrough: a Codex builder and a Claude Code reviewer exchange a review request, a browser finding, a fix, and a recheck on a desktop canvas.
Claude Code
Coding agent
ReviewerSaved role
Check the behavior.
Review conversation
p·
Projects
- Atlas website
- Beacon dashboard
- Cedar docs
Scripted browser check
A place for every part of the work.
Explore the workflow
Place your agents.
Place a builder beside the brief and a reviewer beside the browser. Arrange the canvas around the work.
Start with a brief and two agents.
Read the complete walkthrough
Product concept · Simulated agents and results
- 01
Place your agents.
Place a builder beside the brief and a reviewer beside the browser. Arrange the canvas around the work.
- 02
Give each a role.
A role is a set of instructions you can use again. The builder makes the change; the reviewer checks the result and reports what needs attention.
- 03
Connect the conversation.
Link the agents so the builder can ask for a review and the reviewer can return a finding. Open the connection to see what they said.
- 04
Check the result.
Watch the reviewer enter a search, catch a mismatch, and send the finding back. Then follow the fix through another browser check.
Codex: Builder
Saved role
Implement the requested change. Keep the scope focused. Ask the reviewer to check the result.
Claude Code: Reviewer
Saved role
Check the requested behavior in the browser. Send the builder a concrete finding. Recheck the fix.
The example first returns “No projects found” for “atlas”. After the fix, “atlas” finds “Atlas website”. The input is then cleared, restoring Atlas website, Beacon dashboard, and Cedar docs. Only after both rechecks do the example checks show “Case-insensitive search · Passed” and “Clear search · Passed”.
A plan for your next idea.
An early look at possible plans. Names, features, and pricing are still being worked out.
Canvas
A place to think, arrange, and build.
TBD
Pricing to be confirmedProposed plan
- Notes, tools, and a browser preview
- A canvas you can arrange
- Saved workspace layouts
Builder
Bring a team of agents to your idea.
TBD
Pricing to be confirmedProposed plan
- Everything in Canvas
- Agents with reusable roles
- Connected agent conversations
Studio
Keep the build and the review connected.
TBD
Pricing to be confirmedProposed plan
- Everything in Builder
- Agent-led browser checks
- Review, fix, and recheck workflows
A few things
to know.
Can I download DNP?
Yes. The desktop app for macOS, Windows, and Linux is available from the download link on this page. DNP is in development, so expect frequent updates; the connected-agent features shown in the walkthrough are not in the app yet.
What happens in the preview?
You step through a guided example of two agents making and checking a small change. The example runs no real agents and has no access to your accounts or computer files.
Which agents is DNP being designed for?
Codex, Claude Code, and other AI agents that help you build software. Agent communication and tools are in development; exact compatibility is still to be confirmed.
Does everything run locally?
The workspace is designed to run on your computer, with local terminals and workspace files. Coding agents may still send requests to their model providers.
Will agents be able to use a browser or a phone simulator?
Agent-controlled browsing is a development priority. iOS Simulator and Android emulator access are being explored, with no support or release date confirmed.
What will DNP cost?
Plans and pricing are not confirmed yet. You can explore this preview without a payment or account.
See the conversation
behind the build.
Start with two agents and a simple idea. Step through their roles, conversation, and a check of the result.
The concept runs in the browser, no installation needed. The download gives you the desktop app as it is today.