Your backlog,
their execution
You have ten features in your head. Write them down as tasks, drag one to Development, and a Claude agent starts coding it. Your brain stays clear. Your agents stay busy.
Screenshots
Real captures of the app, added as this feature evolves.
Here, Development is the only column that starts an agent: no other column has a prompt. A QA column is added right after it. A ticket dropped into Development starts an AI coding agent that writes the function the ticket asks for; another ticket dragged into QA simply waits there, and no agent starts.
Now your clients can feed the backlog
Open the same task board to your clients, designers and end users. They report bugs, request features and upvote tickets on a public feedback portal, through an embeddable widget pinned on your production site, or via the Chrome extension that captures any page with a screenshot, the exact CSS selector, the URL and the full DOM context. It's also the shortest path to pull Figma mockups, design references and visual specs straight into a ticket: drop the frame, add a two-line brief, and your agent gets the pixel-perfect context it needs. Non-technical users open a ticket in three clicks, without ever leaving the browser, and you track it all side by side with your own tasks. Every ticket lands directly in your backlog: drag it to Development and an agent running Claude, Codex, GitHub Copilot CLI, Cursor or any of the 20 other agent CLIs AgentsRoom supports ships it. A client-driven backlog, with zero triage tax, no parallel Jira board, no lost email.
Discover the Remote BacklogA kanban board built for AI agents
Not another project manager. A task board where dragging a card to Development literally starts a Claude agent on that task.
Drag to run
Move a task card from TODO to Development. AgentsRoom spawns a Claude agent with the task title as prompt. No buttons, no menus. Just drag.
Four columns, plus your own
TODO for ideas, Development (formerly In Progress) for active agents, Pending for blocked tasks, Done for completed work. Need another stage, like QA or To validate? Add your own columns, each with its own name, color and icon, and give any column its own prompt (analyse, plan, review...): a ticket dropped there goes to the running agent or to a new one. Each column shows a live count of tasks.
Assign agent roles
Assign a specific agent role to each task. Your Security Engineer handles the auth fix, your Frontend Developer builds the new component. Right agent, right task.
Live agent status on cards
Each task card in Development shows the linked agent's status in real time: working, waiting for input, done or idle. You see progress without opening the terminal.
Attach images to tasks
Add screenshots, mockups, or design references directly to task cards. Your agents receive the visual context they need to deliver accurate results.
Parallel task execution
Drag multiple tasks to Development at once. Each one spawns its own agent. Five features in parallel, each with its own terminal, its own status, its own review.
Stored in your repo
Tasks are synced to the cloud and linked to your project. Team members see the same backlog. Progress persists across sessions and machines.
Offload your brain
Stop juggling features in your head. Write them down, prioritize visually, and let agents handle execution. You focus on what matters: reviewing and shipping.

Four columns, unlimited tasks. Drag to execute, track visually.
How it works
Create your tasks
Click + New Task. Give it a title, an optional description, tags, and assign an agent role. Add screenshots or mockups if needed. Your task appears in the TODO column.
Drag to Development
Grab a task card and drop it into the Development column. AgentsRoom immediately spawns a temporary Claude agent with the task title as its prompt. The agent terminal opens automatically.
Watch agents work
Each task card shows a live status indicator linked to its agent. Yellow pulsing means thinking, green means done. Click the card to jump to the agent's terminal and see the full output.
Review and move to Done
When an agent finishes, review its changes in the diff viewer. Satisfied? Drag the task to Done. Need changes? Leave it in Development and give new instructions to the agent.
Repeat at scale
Add ten tasks for your sprint. Drag three to Development in parallel. While agents code, add more tasks you just thought of. Your backlog grows, your agents deliver, your brain stays clear.
Set up the whole board in one click
The first time you open a project's board, AgentsRoom offers a template. One click lays down the columns, and each column comes with its prompt, its agent and a model suited to its job, picked for your own provider. You see what every column will do before you apply, and you can change any of it.
For developers
Classic
TODO · Development · Pending · Done
The essentials: the agent implements the ticket, then sums up what is left to check.
With spec
TODO · Spec · Development · Pending · Done
An architect writes the specification on the ticket with a strong model, without coding. Then a developer implements it and checks every acceptance criterion.
Quality
TODO · Spec · Development · Test · Pending · Done
Spec, development, then a QA agent tests the change against the acceptance criteria before the ticket waits for you.
In Non-dev mode
Everyday requests
TODO · Being handled · Pending · Done
A Personal Assistant handles each request from start to finish, then tells you what to check or decide.
Research and writing
TODO · Research · Being handled · Proofreading · Pending · Done
A Researcher gathers facts and sources, a Writer writes, then a proofreading pass checks spelling, tone and facts.
Your board already has custom columns? The template lists the ones it will remove before you apply. Their tickets go back to TODO, none is deleted, and a column with the same name as one of the template is kept with its tickets.
One ticket, from start to finish
The prompts of the templates keep a single ticket: a spec, a test report or a progress note goes on the ticket itself, in its description or an internal comment, never into sub-tickets unless the ticket asks for them.
A model per column, on your provider
A template stores a level of power, not a model name. On your account it takes the model of that level from your default provider, Claude Code, Codex or another CLI, plus the reasoning effort when that model accepts one. The review screen shows it column by column before anything is applied.
Undo, or restore a prompt
Changed your mind? Undo puts the board back exactly as it was: columns, prompts and the tickets inside them. In a column prompt you edited, Restore the proposed prompt brings back the template's text.
Your own templates
Save this board as a template, for yourself on all your projects and devices, or for the members of this board, team members included. It then appears under Saved templates, next to the built-in ones.
Offered once per project
The template panel opens by itself the first time you open a project's board, once per project, even from another computer. Until then, the Backlog tile of a new project shows a Recommended badge. Later, the template tile above Add column brings the templates back.
Columns in your language, a review gate on demand
The built-in columns are named in the language of the app, and In Progress is now called Development (Being handled in Non-dev mode). Need an approval before Done? New column offers to turn on the In Review column: Done stays locked until a reviewer approves the ticket.
Did you know?
Your users can fill this backlog for you.
1Your users write to you from your public page
2Their requests land in your backlog, votes included
3You drag the ticket: an agent takes care of it
4They are told on their own when it ships
Once the board is ready, a short "Did you know?" animation shows one more idea: with a public backlog, your clients and users send their ideas and bugs, the tickets land on your board, an agent does the work, and everyone who asked is told when it ships. You no longer even need to write the prompts: your clients or users do it for you.
One prompt per column, a pipeline for your agents
Give every column of the backlog its own job. Write a prompt on a column, built-in or one you added, and each ticket dropped there goes to an agent with that instruction: clarify it, plan it, code it, review it, test it. Moving a card becomes a step of your workflow.
Need clarification
Column prompt
Analyse this ticket without coding it, then explain what is unclear.
The agent reads the ticket and the code, then comes back with the questions to settle instead of guessing. Nothing is written to the repository.
Plan
Column prompt
Write the implementation plan for this ticket: files to touch, steps, risks. No code yet.
The plan is ready before the work starts. Drag the ticket on to Development, keep the same agent, and it codes with the plan already in its context.
In Review
Column prompt
Review the changes made for this ticket and list what must be fixed before merging.
Pin a reviewer agent on the column and it checks every ticket that arrives, even when another agent wrote the code.
QA
Column prompt
Run the tests, check the change in the browser and report what fails.
A column you add yourself after In Review, with a QA agent that tests the work before the ticket goes to Done.
Need a mockup to validate with the person who asked, rather than a written analysis? That is Ticket Scoping.
When a ticket is dropped in the column
Open the menu of a column header (Column options), then Column prompt. Write the prompt and pick what happens on each drop. A column that has a prompt shows a Column prompt badge in its header.
- Ask meDefaultEach drop asks whether to keep the running agent, start a new one or just move the ticket.
- Keep the running agentThe agent already working on the ticket receives the prompt and keeps its context. If none is running, a new one starts.
- Start a new agentA fresh agent runs the prompt. The previous one finishes its current turn, then closes.
- Do nothingThe ticket just changes column: no agent starts and none receives the prompt. The prompt stays saved, ready to switch back on.
A column can also pick its own agent, a project agent or a role. With Use this agent even if the ticket has one, on by default, the column's agent takes over; turned off, it only picks up the tickets that have no agent.
Agents move tickets too
When an agent moves a ticket over MCP, the column prompt runs as well, in the background, without stealing your focus. An Ask me column then keeps the running agent, or starts one if none is running.
No endless loops
Moves made by agents start at most 10 column prompts per ticket in 24 hours, so two agents cannot bounce a ticket forever. Change it in Settings > Backlog > Automatic column prompts per ticket, for every project or for one; 0 means unlimited. Your own moves are never limited.
Phone, every CLI, every computer
Move a ticket from the mobile app and the same choice appears in a bottom sheet, then your desktop runs it; prompts are written on the desktop. It works with Claude Code, Codex and the other agent CLIs, on macOS, Windows and Linux, and the prompts are stored on your account, so all your computers share them.
Epics, tasks and subtasks, on the links you already have
Large work stops being a pile of loose cards. An epic holds its tasks, a task holds its subtasks, and the board shows how far each one is, what blocks it and what an agent should do next.
An Epics view, column by column
Switch the board to Epics: every epic with its progress and what comes next, and its tickets in the columns where they stand. A click on its name opens it.
Subtasks fold into 3/5
A subtask folds under its task while they share a column, and shows as its own card as soon as an agent takes it.
Blocked by #12, on the card
Set a dependency from the card's menu. A loop is refused, and a blocked ticket waits in Pending until its blocker closes.
Agents pick the right ticket
Start next skips epics, parents and blocked tickets. An agent receives the epic's goal and how far its siblings are.
Start epic, with a limit
Launch the ready tickets of an epic in dependency order, up to the limit you set, each in its own worktree if you want.
One epic tonight, then stop
Run an epic to its end, now or at a set time. Tickets start as others close, its tests and security checks come last, and nothing else starts.
Epics in order, with a pause
One epic can wait for another: its tickets stay put until the first is closed. Pause holds an epic the same way, and agents at work finish their ticket.
Tasks that execute themselves
In Jira, Linear, or Notion, tasks are static cards. You read the task, then switch to your IDE, then start coding. In AgentsRoom, dragging a task to Development spawns an agent that starts coding immediately. The board is not just for tracking: it is the trigger.
Your tickets from GitHub Issues, GitLab Issues, JIRA Cloud, Notion, Linear, Azure DevOps, YouTrack and Odoo
Part of the work already lives in another tool. Pick a source, connect it, choose what to bring in, preview, import. Then replay the same import whenever you want.
Eight sources, one board
GitHub Issues with a repository, a state and labels. JIRA Cloud with a project or a JQL query. Notion with a database shared with the integration. Linear with one or several workspaces, each with its own API key, filtered on what is assigned to you, by team, project and status. Azure DevOps with a project or a WIQL query and a personal access token. GitLab Issues with a project, a state and labels, on gitlab.com or your own instance. YouTrack with a project or any YouTrack search, on YouTrack Cloud or your own instance. Odoo with a CSV file of tasks exported from Odoo Project, no connection and no password. You preview the result before anything lands on the board, and a search field finds the one issue you already know the key of. Only need that one? Paste its reference straight into the TODO column and it comes in on its own.
Replay the import, no duplicates
Running the same import again updates the tickets already imported instead of creating copies. Title, description, type, labels and the screenshots attached to the issue are rewritten from the source on every pass, so the two boards stop drifting apart.
Your own work is never overwritten
The assigned agent, the internal notes, the order on the board and the links between tickets survive every refresh. A ticket already in progress or done here never drops back to TODO, and an issue that leaves the query keeps its ticket.
Read only by default, two-way sync as an option
By default the import only reads: nothing is pushed back to GitHub, GitLab, JIRA, Notion, Linear, Azure DevOps or YouTrack. Turn on the two-way sync for a project and column moves, title, description, comments, new tickets and the pull request link go back to the tool, while a change made there moves the ticket without ever starting an agent.
The import token stays on your machine
Your import token is encrypted locally and never reaches our servers. Only the query is saved on your account, so the refresh follows the project from one machine to the next. The optional two-way sync uses a separate write token, kept encrypted on our servers so it runs with your computer off.
Every ticket says where it comes from
An imported ticket shows its source, links back to the original issue one click away, and can be brought back in step on its own without replaying the whole query. Setup happens on the desktop; from the phone you can refresh the whole import or bring in a single ticket, and both run on your computer.
A board for ideas that have no project yet
Some ideas are not ready for a repository. Drop them in the idea box, a kanban attached to your account rather than to a project, and move each ticket to the right board the day it becomes real work.
Tickets without a project
The idea box is a regular backlog board with one difference: it belongs to you, not to a repository. Write the idea down as a ticket, with a description, tags and screenshots, and stop carrying it in your head. It is synced on your account like every other board, and the mobile app opens it from the lightbulb icon in the projects list.
Move a ticket to another board
From the card menu, or with several tickets selected, choose Move to another board and pick a project. The ticket travels as is, with its description, its attachments and its history, and lands at the top of its column. It also works between two projects, for the ticket that was filed on the wrong board.
Starting an idea means moving it
The idea box never starts an agent: there is no project to run it in. Drag an idea to Development, or press Start, and AgentsRoom asks which project to move it to. The board follows the ticket, and the agent starts there, on real code.
Change board without closing it
The project name at the top of the board is a menu: the idea box first, then the projects open in this session, then the others. Pick one and the board switches to it, without closing the window and clicking through the sidebar.
How does dragging a task to Development start an agent?
Can I run multiple tasks in parallel?
What happens if an agent finishes but I am not satisfied?
Are tasks shared across my devices?
Can I attach images or screenshots to tasks?
Can I merge duplicate tickets?
Can one ticket run on a different provider or model from the rest?
Can a ticket wait for another one to be finished?
Where do I keep ideas that do not belong to any project yet?
Can I move a ticket to another project?
Does a ticket keep a history of its versions?
Can my agents leave notes on a ticket for the rest of the team?
How do I see which tickets will run in their own worktree?
Can a ticket wait for a decision instead of another ticket?
Can a kanban column start an AI agent with its own prompt?
Do column prompts run when an agent moves a ticket, and what stops a loop?
How do I find a ticket an agent is talking about?
Can I set up the backlog board from a template?
How do I organize a large feature into epics, tasks and subtasks?
What does an epic show me?
How do agents work through an epic?
Can AI break an epic down, and can a ticket carry acceptance criteria?
Can I use the backlog without an account, or behind a company firewall?
Can I schedule a ticket to start later, on a specific computer?
Can I run one epic overnight and have it stop there?
Can an epic be tested and checked for security before the next one starts?
Can I move a ticket to another column without going back to the board?
You may also like
Ticket scoping
Before an agent builds a feedback ticket, scope it: a Product Manager AI agent frames the request into a mockup faithful to your product and validates it with the reporter.
Public remote backlog
Let clients and teammates file feedback tickets that land straight in the backlog your agents execute against.
Linear import
Import your Linear issues into AgentsRoom and hand each one to Claude Code, Codex or any agent CLI. Several workspaces, assigned to me, teams, statuses. Read only by default, two-way sync optional.
Git Worktrees
Give each AI agent or ticket its own git worktree, ready to build (dependencies installed, .env copied), then review, commit and merge it from one window.
Further reading
A Feedback Board for AI Agents: Let Your Users Write the Prompt
Feedback tools collect requests. None of them can build one. When the board your users write into is the same board your coding agents execute from, the rewriting step disappears.
How to Run 3 to 8 Coding Agents in Parallel Without Losing Track
Launching multiple Claude Code or Codex agents at once is easy. Keeping control is where it falls apart. Here's the method that actually works.
Agents Write the Code Now. Here Is What the Developer's Job Became.
Writing code was one link in a six-link chain, and it is the link agents took. The other five got heavier. A walkthrough of the job that is left: hearing what to build, deciding, briefing, running, reviewing, shipping.
Turn your backlog into running agents
Download AgentsRoom and start dragging tasks to execution. Free for up to 2 projects.
Companion app: monitor your agents on the go
Bring your own: Claude, Codex, Antigravity CLI, or other AI provider.
Push bugs and requests straight to your public backlog.













