Three months ago
T-042Done
Token rotation
Claude Code Claude
Delivery and decision recorded under Auth.
Your project,beyond the chat.
Connect your agents, their tickets and the decisions that matter.
Your runtimes. Your models. Your project.
One brief → tickets → reviewed deliveriesFollow the context
The team at work.
Explore agents, tickets and deliveries in the demo.
Team board
See the full demo
The full interactive demo runs at demo.agents-deck.com.
One point of contact: the Team Leader.
Ask a question about Agents Deck.
Team LeaderAI assistant
This product guide does not launch tickets in the demo.
Who works. Who delivers. Who accepts.
Each collaborator has a scope and skills to apply concretely. The Team Leader verifies that application before accepting: a non-compliant delivery is refused.
EXAMPLE · SSO LOGINDatabaseSchemas & migrations
T-001Identity schema
DoneWork accepted
Accepted by Team Leader
BackendAPI & authentication
T-002SSO login
WorkingIn progress
Pending
SecurityPermissions & risks
T-003Access review
ReviewDelivery submitted
Awaiting Team Leader review
You set the scope. The Team Leader organizes and reviews.
Animated example
You
Add SSO authentication.
Keep the current loginGet the plan approved
Team Leader
T-001Database
T-002Backend
T-003Security
Collaborator → Team Leader
CollaboratorReview
Team LeaderDone
Inspect the result + run checks
ONE REQUEST. AN ENTIRE NETWORK.
Start with the Team Leader.
In Codex, define the request and its boundaries together.
TEAM LEADER / DISCUSSIONFollow the ticket as you scroll
SessionTickets / groupsAgentsTeam LeadersTeams / projectsShared trackingInternal tracking
Illustrative scenario · tracking shared across projects · deliveries reviewed by Team LeadersKeep scrolling ↓
History that helps the next ticket.
The subjectAUTH-01
Shared tracking, links to past work.
The recordT-042
Accepted delivery and Team Leader decision.
What’s nextT-184
A new session consults this context.
The collaborator stays. Sessions change.
Backend keeps its role and history. For each ticket, the Team Leader chooses the execution according to your model policy.
PERSISTENT COLLABORATORBackend
API · Auth · Tests
Same role, same skills
History consulted
New mission
T-184New session
Session revocation
GPT
The relevant context is read for this execution.
Illustrative choice before launch. The model can be required, recommended or freely chosen. Available operations depend on the runtime.
What you can organize with Agents Deck.
The tools below already delegate work, retain sessions or orchestrate agents. Agents Deck connects that work to responsibilities, deliveries and a traceable project history.
Built inAssembleNot established
| Your workflow |  AgentsDeck | ||||||||
|---|---|---|---|---|---|---|---|---|---|
Built inRole · scope · skills | |||||||||
Built inTicket + session + delivery | |||||||||
Built inReview → Team Leader approval → Done | |||||||||
Built inSubject → history → tickets | |||||||||
Built inRuntime chosen per mission | |||||||||
Built inEncrypted on your machine |
UNDERSTAND THE DIFFERENCE
Agents Deck
Understand a decision three months later.
A tracking record describes the subject and links to tickets, deliveries and recorded decisions. The collaborator follows those links to reconstruct the relevant context. Structured memory is available, with an optional local context search.
CodexNative delegation
What is already provided
Inspectable threads and worktrees subject to retention policies.
Read the documentationConcrete exampleAuth → T-042, completed in June → token decision → constraint reused in T-184.
“Assemble”: configuration or development needed for this workflow. “Not established”: unconfirmed in the reviewed sources, not necessarily impossible.
Details, limits and sources
Reference review: September 28, 2026. Agents Deck: creator’s product brief. Alternatives: official documentation. Features compared, not performance.
Agents DeckOne Team Leader per project. No automatic restart. The app is macOS-only; Windows and Linux use the command line.
Claude Agent Teams
- Organization
- Lead, teammates and shared tasks with dependencies. Direct agent messaging.
- Tools & models
- Claude Code sessions; reusable subagent roles.
- Memory
- Persisted tasks, project memory and role definitions.
- Acceptance
- Plan approval and hook controls. Business acceptance policy needs configuration.
- Execution & data
- Sessions in your environment; model processing depends on the configured provider.
- Known limits
- Experimental. Resuming the lead does not restore running teammates.
- Costs to plan for
- Subscription or model usage. Multiple contexts and coordination add consumption.
Claude Projects
- Organization
- A coordinator delegates to threads and checks results.
- Tools & models
- The redesigned Projects experience in Claude; limited access in the announcement.
- Memory
- Shared memory and an artifact library.
- Acceptance
- Coordinator review; acceptance policy needs evaluation.
- Execution & data
- Cloud threads. Local execution announced, not shipped in the announcement reviewed.
- Known limits
- Limited beta. Retention and export were not established in the review.
- Costs to plan for
- Announced for selected Pro/Max subscribers; eligibility and limits need checking.
Codex
- Organization
- The main agent delegates to subagents and combines their results.
- Tools & models
- Configurable agents in Codex; application worktrees.
- Memory
- Inspectable threads and worktrees subject to retention policies.
- Acceptance
- Sandbox and operation approvals. A business Review/Done workflow needs assembly.
- Execution & data
- Local or cloud depending on mode; configurable permissions.
- Known limits
- The cross-project business registry described for Agents-Deck needs building.
- Costs to plan for
- ChatGPT plans as of 09/28/2026: Free $0, Go $8/month, Plus $20/month, Pro from $100/month, or separate API usage.
Conductor
- Organization
- An external manager can create, supervise and redirect workspaces through API/MCP.
- Tools & models
- Multiple runtimes: Claude Code, Codex, Cursor and OpenCode.
- Memory
- Identified sessions, messages and transcripts linked to a workspace.
- Acceptance
- An acceptance workflow can be composed through API/MCP.
- Execution & data
- Local application or cloud; session exposure differs by mode.
- Known limits
- Beta API. Business responsibilities and tracking need arranging around workspaces.
- Costs to plan for
- As of 09/28/2026: free local; Pro $50/month; Teams $60/user/month by invitation; Enterprise on request.
Vibe Kanban
- Organization
- Work management, workspaces and sessions controlled through local MCP.
- Tools & models
- Multiple executors in a development-oriented workflow.
- Memory
- Documented sessions and execution tracking; check persistence in the installed version.
- Acceptance
- A business acceptance workflow can be composed around MCP.
- Execution & data
- Public code and local use; former remote services are in transition.
- Known limits
- Bloop announced its closure. Community maintenance and migrations need review.
- Costs to plan for
- Open-source continuity announced. Infrastructure, models and maintenance still cost money.
Cursor
- Organization
- The parent agent delegates to specialists in the editor, CLI or cloud.
- Tools & models
- Configurable models and tools within the Cursor ecosystem.
- Memory
- Reusable specialist definitions and resumption by identifier.
- Acceptance
- Hooks observe or control stages; business rules need connecting.
- Execution & data
- Local or cloud execution. Self-hosted machines still use Cursor’s agent loop and inference.
- Known limits
- Multiple models do not mean orchestration of multiple independent CLIs.
- Costs to plan for
- As of 09/28/2026: free Hobby; individual from $20/month; Teams from $40/user/month; Enterprise on request.
CrewAI
- Organization
- Hierarchical process: a manager assigns tasks and validates results.
- Tools & models
- Configurable model providers and models per agent.
- Memory
- Configurable memory and retrieval during tasks.
- Acceptance
- Function or LLM guardrails, with feedback and retries.
- Execution & data
- Deploy the application and infrastructure for your architecture.
- Known limits
- UI, permissions, tickets and encrypted storage need design and maintenance.
- Costs to plan for
- Development, operations, models and optional managed services.
LangGraph
- Organization
- Stateful graph: hierarchy and transitions defined in your application.
- Tools & models
- Models and tools integrated for your chosen architecture.
- Memory
- Per-thread checkpoints; business memory depends on the schema and stores.
- Acceptance
- Interrupt, receive an external decision, then resume from saved state.
- Execution & data
- Infrastructure, storage and access are the integrator’s responsibility.
- Known limits
- Build tickets, roles, UI and observability. More control requires more integration.
- Costs to plan for
- Development, infrastructure, models and optional managed services.
Install the app. Sign in once.
Create your account, install the Agents-Deck app on your Mac and sign in. It links this machine, then sets up your project and your AI tools.
Never run npx agents-deck. That npm name belongs to an unrelated third-party package, also published as ccdeck and agent-dag. The only legitimate downloads are the Mac app from the console and https://get.agents-deck.com/install.sh, which verifies the binary's SHA-256 checksum before installing it.
01
Create your account
in the browserSign up at console.agents-deck.com. The Connection page offers the Mac app for Apple silicon and Intel Macs.
Open console.agents-deck.com02
Install the app and sign in
on your MacInstall it, open it and sign in with your account. It links this machine and starts the background service on its own.
03
Create a project, connect your AI tools
in the app“New project…” picks your folder. “Connect my AI tools” sets up Claude Code and Codex. Then follow the three steps to start your Team Leader session.
Windows, Linux or a server
Install the command-line companion, then sign in with a machine token created on the Connection page.
Linux (and macOS via the command line)
Windows (PowerShell)
Start with one project.
More room with Pro.
Use your own models and keys. Model usage is billed separately.
Free
Available nowOrganize your first team.
0€with no time limit
Project1
Collaborators3+ one Team Leader
No credit card required
- Project dashboard
- Unlimited tickets and tracking
- Community support
Pro
WaitlistWork across multiple projects.
19€/ user / month
ProjectsUnlimited
CollaboratorsUnlimited
Announced plan · no payment to join
- Everything included in Free
- Multi-user access and per-project roles
- Email support planned
Included in both plans
CLI & MCPTickets & memoryReview before DoneEncrypted on your machine
Enterprise Custom pricing
Need SSO, shared access or help with setup?
Frequently asked questions.
Who decides to create a ticket? +
You scope the need in your Team Leader's session: objective, scenarios, security and acceptance criteria. It proposes a breakdown into tickets and collaborators, then transmits them once the plan is approved for that mission; broader autonomy can be authorized under your procedure.
Can the Team Leader act alone, without human validation at every step? +
Within the scope you authorize, yes: it can chain analysis, corrective tickets, follow-ups and integration. If it gets something wrong, a collaborator can flag disagreement; the Team Leader re-examines the approach, arbitrates or cancels the ticket in question.
Do I need to talk to each agent separately? +
The normal workflow starts in your Team Leader session. It prepares tickets, launches collaborators and resolves blockers. You can also resume a collaborator’s session to inspect its work.
Which tools and models are supported? +
Managed launches are verified for Claude and Codex. The Team Leader and collaborators can use other tools depending on your setup, but connecting another tool over MCP is not the same as verified launch, resume or measurement support.
Are required skills actually applied, or just declared? +
They must be applied concretely in the delivered work, not just mentioned. The Team Leader checks this application before accepting; a non-compliant delivery is refused and sent back for correction.
Who accepts a delivery? +
The collaborator submits to Review with its report and required skill validations. The Team Leader inspects and verifies the result before deciding. The collaborator cannot mark its own ticket Done.
Does project memory depend on one session? +
No. Tickets, deliveries, decisions, skills and tracking tables make up the project memory. A local search can retrieve that context, with optional embeddings computed on your machine; it does not cover the entire history or every past exchange.
Can several people collaborate on a project? +
Yes, through invitations and roles managed in the console: adding, changing the role of, or removing a member. A project keeps a single active AI Team Leader; several humans can take part according to their rights.
Can I self-host Agents-Deck? +
The offering is a SaaS paired with a local companion that runs on your machine and encrypts your content before it syncs. There is no fully self-hosted offering and no open-source status for this product.
What can the service read about my projects? +
Your tickets, deliveries, messages and tracking are encrypted client-side: the service cannot read them. It does keep readable metadata: your account, project, folder and device names, states, relationships and activity dates.
What happens if I lose my keys? +
If you configured a backup protected by a secret phrase or a recovery code, a new machine can retrieve the key without the old one. Without any of these means, access to encrypted content cannot be restored.
What do the Insights dashboards show? +
Project- and workspace-level dashboards with configurable widgets: activity, tickets, durations, models, agents, tokens and costs. Data can be complete, partial or unmeasured; an unknown cost is never counted as zero.
Do I need an external Git host like GitHub? +
No. The companion keeps a local Git history on your machine; no external account or remote is required to deliver a ticket.
Based on the product reference dated September 28, 2026.