Skip to content
Download DNP
Agents

Agents

Claude Code and Codex as local agent seats, each with a role, a task and a status you can read.

Claude Code and Codex.Together on one canvas.

Give each seat a role and a task. One can map the change, another can build it, a third can review it.

Workflow diagram: role-pipelineIllustrative role pipeline: brief, agent tasks, review and your decisionBriefContractDiffReportBriefUser requirementArchitecture & testsArchitect (Agent)Plan the changeMap contracts & slicesBuilder (Agent)ImplementWrite pure reducers & filesReviewerCheck the briefValidate contractsYou decideHuman reviewMerge or iterate
Workflow diagram: role-pipeline (mobile)Illustrative role pipeline: brief, agent tasks, review and your decisionBriefPlanDiffReportBriefTask requirementsArchitect (Agent)Plan the change · Claude CodeBuilder (Agent)Implement · Codex CLIReviewerCheck the briefYou decideHuman approval
Illustrative role pipeline: brief, agent tasks, review and your decision

Roles describe the task. They are not separate agent products.

Inside a seat.Role, task, status.

Choose a supported CLI, describe the role and the task, then follow the status.

Workflow diagram: seat-anatomyAgent seat anatomy: CLI, role, task, status and an attached local terminalCLIRoleTaskStatusShell01 / CLIClaude Code or CodexSigned in on your Mac02 / RoleArchitect, Builder, or ReviewerAccountable seat definitionAgent Seat AnatomyRole: BuilderTask: Implement the briefStatus: Running (pid 4082)CLI: Claude Code03 / TaskSingle bounded jobFrom the workspace brief04 / StatusIdle / Running / WaitingVisible at a glanceTerminal childAttached workspace terminalExecutes tool & build steps
Workflow diagram: seat-anatomy (mobile)Agent seat anatomy: CLI, role, task, status and an attached local terminalAttached shellAgent Seat AnatomyCLI: Claude Code or CodexRole: BuilderTask: Implement briefStatus: RunningAttached TerminalDirect local shellFour seat attributes1. CLI: Your existing accounts2. Role: Defined accountability3. Task: Verifiable brief4. Status: Visible at a glance
Agent seat anatomy: CLI, role, task, status and an attached local terminal

The task and the available output stay in view as the work progresses.

Review the change.Then decide.

A Reviewer seat compares the change with the brief and the test output. You make the final call.

Reviewer feedback is available on Free today. Enforced reviewer policy gates are planned for Pro.

Your CLI.Your login.

Sign in to Claude Code or Codex on your Mac. Model access and billing stay with your provider.

Workflow diagram: byo-cliBYO CLI: your existing login feeds a DNP agent slot; billing stays directAuthAuthRuns CLIRuns CLIDirect usageExisting CLI loginclaude / codex loginSigned in on your MacClaude Code CLILocal CLI binaryAnthropic subscriptionCodex CLILocal CLI binaryOpenAI subscriptionDNP Agent SlotVisual orchestrationNo token markup · No proxyYour provider planPro / Team / API creditsBilled directly by providerModel ProviderAnthropic / OpenAIDirect network exit
Workflow diagram: byo-cli (mobile)BYO CLI: your existing login feeds a DNP agent slot; billing stays directAuthRuns CLIDirect usageYour existing CLI loginSigned in on your MacClaude Code & CodexLocal CLI execution binariesDNP Agent SlotVisual canvas · Zero token markupYour provider planDirect model usage billingModel ProviderAnthropic / OpenAI directly
BYO CLI: your existing login feeds a DNP agent slot; billing stays direct

DNP does not sell tokens and does not proxy agent requests.

Seats ask first.You approve.

Read an agent’s request before it continues. The action and your decision stay in a local transcript.

Workflow diagram: approval-transcriptApprovals: an agent request waits for your decision, then runs locallyRun requestLogsApproveDenyAuditAgentClaude Code or CodexWants to run a terminal stepTranscript artifactRequest: pnpm test:uiDecision: Approved by userOutput: Exit code 0Approval requestRun project command?pnpm build --verboseApproveCommand executesRecorded in transcriptDenyCommand blockedFeedback sent to agentYou (Human)Full control overlocal execution
Workflow diagram: approval-transcript (mobile)Approvals: an agent request waits for your decision, then runs locallyRequestApproveDenyAgentWants to run commandApproval requestRun project command?pnpm testApproveExecutes commandDenyBlocks commandTranscript artifactRequest / Decision / OutputLogged to diskYou decideTransparent approvals
Approvals: an agent request waits for your decision, then runs locally

Approvals and local transcripts are included in DNP Free.

Questions.Short answers.

Supported agents, local work and provider accounts.

Which agents is DNP designed for?

Claude Code and Codex are supported today. DNP runs the CLIs you install and sign in to on your Mac; each agent gets a role and a task on the canvas.

Does everything run locally?

Your canvas, project files and terminal work stay on your machine. The agent CLIs send model requests to their providers using your own account.

How does the review loop work?

Add a Reviewer seat to compare the change with your brief and checks. Read its feedback, then decide what to merge. Enforced gates that block a run until checks pass are planned for Pro and are not available yet.

Will agents be able to use a browser or a phone simulator?

The Browser block is planned and is not available yet. iOS Simulator and Android emulator support are not available, and no release date is confirmed.

Do I need my own AI subscription?

Yes. Sign in to Claude Code or Codex with your own provider account. DNP supplies the canvas, not model access, and does not sell model usage.

Does DNP charge for tokens?

No. Model usage is between you and your provider. DNP does not sell or mark up tokens.

Start your first canvas.

Download DNP for macOS on Apple silicon, open a canvas and give a seat its first task.