AgentDeck

Local-first AI agent workspace

Every AI agent. One governed workspace.

Keep the personal plans and team seats your people already use. Add enterprise model routes when you need them. AgentDeck brings the work into one local-first desktop workspace—with secure mobile continuity and organization policy built in.

See the platform Get the desktop beta

No token resale or model markup. Provider plans, seats, terms and limits remain with their owner.

AgentDeck Live workspace Connected
8open sessions
5working
2need you
3projects
Work in motionright now
AgentDeckFix PWA reconnect loop
Codexediting
Maidepot APIDeploy staging worker
Claudeapproval
Website relaunchBuild pricing comparison
Geminiresearching
AgentDeckSecurity regression suite
Claudefinished
Personal plans and team seats Cloud routes without lock-in Local work, paired mobile control Policy without reading conversations

WHY AGENTDECK

Choose the intelligence. Keep control of the work.

Models, clouds and subscription plans will keep changing. AgentDeck gives the work a durable operating layer across providers, accounts, projects, devices and teams.

Vendor freedom

Change the model, not the workspace.

Use Claude, ChatGPT and Codex, Gemini, or organization-managed routes through AWS, Google Cloud, Azure and compatible gateways.

Projects and operating habits outlive a provider choice.
Bring your own access

Keep the plans and seats you already have.

Individuals sign in to their own plans. Team members sign in to provider accounts assigned by their organization. Credentials stay on their computer.

No pooled logins. No central vault of provider tokens.
Govern without surveillance

Manage access without reading the work.

Enterprise administrators manage seats, devices, policy, usage and audit metadata—not prompts, responses, files or command content.

Control the operating boundary, not the conversation.
Work from anywhere

The computer works. Decisions find you.

Monitor progress, answer questions and review approvals from a deliberately paired phone while tools and project files stay on the computer.

One work state across desktop and mobile.

ONE OPERATING LAYER

Accounts stay with their owner. Work comes together here.

AgentDeck connects provider access, local tools and human decisions without taking ownership of the underlying subscription.

01 / CHOOSE THE ROUTE

Your AI access

ClaudeChatGPTCodexGemini

Personal plan or team seat: each member signs in directly with the provider account assigned to them.

02 / RUN THE COMPUTER

AgentDeck

  • Projectsfolders and context
  • Sessionsconversation and state
  • Actionsfiles, tools and terminals
  • Decisionsquestions and approvals
03 / GOVERN THE BOUNDARY

Personal + Enterprise

USERpolicyORG

Personal settings remain flexible. Signed organization policy sets the ceiling, and the stricter rule wins.

Subscription ownershipIndividual or organization

The provider plan or team seat remains where it is purchased and administered.

Provider sign-inEach member, directly

Members authenticate on their own computer. AgentDeck does not distribute shared provider credentials.

AgentDeck governanceOrganization policy

Enterprise controls AgentDeck seats, devices, allowed capabilities, usage and audit metadata.

Important boundary: removing an AgentDeck seat prevents new organization sessions in AgentDeck. Revoking the member’s upstream provider account or team seat still happens in that provider’s admin console.

SEE IT WORK

Your computer keeps working. You stay in control from your phone.

Desktop runs the tools and files. Your paired phone keeps the important decisions within reach.

Maidepot APIDeploy staging worker
C

I finished the tests. The staging deploy needs your approval before I continue.

Approval required gcloud run deploy maidepot-api-staging External change · reversible deployment · staging project
Waiting for your decision
12:49AgentDeck●
MAIDEPOT API

Deploy staging worker

Claude needs approval to make an external change.

Commandgcloud run deploy …
RiskExternal deployment
01

Leave the desk, not the loop.

Review the exact command, risk and destination from your phone. Approve once, deny or answer the agent’s question without opening a laptop.

KEEP THE MOMENTUM

A rate limit should change the account, not erase the work.

AgentDeck can continue a stopped session on another account of the same engine. The project folder, conversation and unfinished task remain one session in your workspace.

  • No copy-pasting a giant context prompt
  • No duplicate session clutter
  • No repeating work that already finished

Handoff uses another account you are authorized to use on the same engine. It does not alter or evade the provider’s limits, terms or access controls.

AgentDeckPrompt-injection hardening
same session
context→
Resumed from the unfinished security testworking

YOUR WORK, NOT THEIR BRANDS

Organize around projects. Use whichever agent gets the job done.

▰
AgentDeckC:\dev\agent-deck
4
CodexFix PWA reconnectworking
ClaudeSecurity reviewfinished
GeminiCompare onboardingwaiting
▰
Maidepot APIC:\dev\maidepot-api
2
ClaudeDeploy stagingapproval
CodexRepair queue workerworking
＋
Start something newChoose a folder or let AgentDeck create one
What do you want to build?↑

SECURITY BY DATA BOUNDARY

Protect the context before it becomes a prompt.

AgentDeck puts security checks in the local execution path, where project context, provider credentials and action authority can be separated. Organization controls receive operating metadata—not a copy of the work.

01 / LOCAL INPUT Prompt + project context files and credentials remain on the computer
02 / SECURITY GATE PII · context · injection inspect, pseudonymize, block or require a decision
03 / APPROVED ROUTE Selected AI provider provider policy and region still apply
BUNDLED · LOCAL

PII protection without a model download

Bundled multilingual NER and deterministic checks can pseudonymize supported names, contact details, identifiers, card data and secrets before a request, then restore the response locally.

EXPLICIT · REVIEWABLE

Context masking that does not pretend to read minds

Teams define organization, product and customer terms to protect. Conservative patterns can cover domains and infrastructure identifiers. Stronger path and repository masking remains an explicit choice because it can reduce agent accuracy.

UNTRUSTED BY DEFAULT

Prompt injection is not execution authority

Retrieved content is treated as data, not permission. Folder scopes, destination checks, risk cues and explicit approvals constrain what an injected instruction can turn into.

FAIL CLOSED

No silent downgrade when a required check fails

When organization policy requires a security check and that check cannot complete, the protected turn is blocked instead of quietly sending uninspected content.

What “local-first” means here

AgentDeck does not send prompts, responses, files, command content, local paths or provider credentials to its central control plane. Content you ask an agent to process is still sent to the selected AI provider or connector, under that service’s account, region and data terms.

AGENTDECK ENTERPRISE

One workspace for people. One operating boundary for the organization.

Enterprise adds the control layer needed to adopt agentic work across a team without centralizing everyone’s provider login or exposing their conversations to administrators.

  • Roles, invitations, AgentDeck seats and approved devices
  • Signed policy for engines, permissions, security checks, MCP and remote shell
  • Usage, cost and security-event metadata in one view
  • AWS Bedrock, Google Vertex AI, Azure AI Foundry and OpenRouter-compatible gateway routes
Explore Enterprise Talk to us
AgentDeck EnterpriseOrganization overviewPolicy active
Assigned seats42of 50
Active devices382 need review
Policy coverage100%revision 18
Security checksPII required · context reviewEnforced
Model routesProvider seats + company cloud4 allowed
Device postureSignals, version and policy state2 review
Conversation contentNot collected by the control planePrivate
Audit metadataUsageSeatsDevicesPolicies
THE ORGANIZATION CAN SEE

Seat assignment, provider route, model, timestamps, usage, cost estimates, device posture, policy revision and enforcement events.

THE ORGANIZATION CANNOT SEE

Prompt and response text, conversation history, files, command content, provider credentials, profile directories or local project paths.

QUICK ANSWERS

Before you bring your accounts into one workspace.

What is AgentDeck?

AgentDeck is a local-first desktop workspace for running and monitoring AI agent work across projects, with secure continuation and approvals from a paired phone.

Does AgentDeck replace Claude, ChatGPT or Gemini subscriptions?

No. Keep the personal plans or team seats you already use and sign in directly with each provider. AgentDeck organizes the work without pooling passwords, cookies or provider tokens.

Can AgentDeck administrators read prompts, responses or files?

The control plane does not collect prompts, responses, files, command content, local paths or provider credentials. Content selected for an AI task still goes to that provider under its account, region and data terms.

Can I continue computer work from my phone?

Yes. After deliberate device pairing, your phone can show progress, questions, destinations and approval requests while tools and project files remain on the computer.

Does AgentDeck bypass provider usage limits?

No. Provider terms and limits still apply to each account. AgentDeck coordinates independently owned subscriptions and sessions; it does not circumvent an upstream provider limit.

MAKE AI ACCESS OPERABLE

Use the AI your people prefer. Govern the work your organization owns.

Start with an existing personal plan or team seat. Add organization policy and enterprise model routes when the work demands them.

Get the desktop beta See Enterprise