Skip to content
PortBay
Comparison

PortBayvsAgent Kanban

Agent Kanban makes an argument about PortBay's whole category, so let's start there. Its site says local orchestrators "run agents on your laptop — the board dies when the lid closes", and offers a hosted queue instead: your agents poll /tickets/ready from a cron job, a CI runner or your machine, claim work, comment the result and move it to In Review, while you watch the board from anywhere. That criticism is fair, and this page does not pretend otherwise — PortBay's board lives in a desktop app on one Mac, and when that Mac sleeps, so does the board. What the trade buys is the other half of the loop. Agent Kanban never starts an agent; something else has to be running for the queue to matter. PortBay is that something: move a card to Todo and it launches your agent CLI in its own git worktree, inside a macOS Seatbelt profile, against a project it has already brought up with a pinned runtime, a database, trusted HTTPS and captured mail. One is where the work is written down. The other is where the work happens.

PortBay vs Agent Kanban comparison page.
The short answer

Which one is right for you

Pick PortBay if

You want the full stack, open source.

The hard part is not tracking the work, it is running it. You want the card to start the agent, in a throwaway worktree, under a macOS Seatbelt profile with a network policy, against an application that is already up — so it can apply a migration, load the page over HTTPS and read the mail it just sent. You want tasks as markdown files inside the repository, diffable in a pull request and readable offline, rather than rows in a hosted database. And you want the runtime, seven database engines, the .test domain and the mail catcher to come with the board rather than be your problem.

Stick with Agent Kanban if

It already fits your workflow.

Your agents do not all live on one Mac, and their point about the lid stands. Agent Kanban is a URL, so a 3am cron job, a CI runner, a Linux box and your phone all read the same queue, and a purpose-built /tickets/ready endpoint exists precisely so an unattended agent can poll it. Its workflow is finer-grained than PortBay's — Backlog, Refinement, Ready, In Progress, In Review, Done, with Low/Medium/High/Critical priorities and an activity log — and everything is both an MCP tool and a plain REST endpoint with a generated OpenAPI spec. There is a real free tier (1 project, 50 tickets a month, 1 agent key) and Pro is $5/month, which is cheaper than PortBay. If you are on Windows, if your agents run mostly in CI, or if you want a queue and nothing else, pick Agent Kanban.

Where PortBay and Agent Kanban actually differ

The table below has the row-by-row detail. These are the three shapes it cannot draw: the job you already use one of them for, the job only one of them does, and what survives the move.

Both run your sites

PortBay and Agent Kanban overlap on the everyday job: a local site on a real domain, over HTTPS, with the services it needs. If that is all you need, either will do it.

Only one dispatches agents

Move a card to To Do and PortBay launches Claude Code, Codex or Cursor against a project that is already running. Agent Kanban has no board to dispatch from.

Switching keeps your projects

Point PortBay at the folders you already have. They stay where they are, .test domains are re-issued on the first run, and databases are imported rather than rebuilt.

The detail

Feature by feature

Agent Kanban rows come from Agent Kanban’s own site and docs, on the date below. PortBay rows come from the app’s source. We mark partial support honestly — including where the other side wins.

Feature
PortBay
Agent Kanban
What each tool is
Primary job
Local dev env + agent board
Hosted queue agents poll
Launches the agent for you
Yes — dispatched from a card
No — agents poll in
Agent Kanban documents the loop as the agent's job: poll /tickets/ready on a schedule, claim the ticket, do the work, comment back. Something else — your terminal, your cron entry, your CI job — has to be running that agent. PortBay starts it from the card.
Board survives the laptop closing
No — desktop app
Yes — hosted, always up
Their argument, conceded. A hosted queue keeps working while your Mac sleeps; PortBay's board is in a desktop app and does not. If your agents run unattended overnight, this row is the whole comparison.
Reachable from cron, CI, a phone
Local to one Mac
Yes — any HTTP client
Works with the network off
Yes — files on disk
No — hosted service
The same trade, read the other way. PortBay's cards are files on disk, so the board opens on a plane; a hosted queue does not.
Platforms
macOS (ARM + Intel), Linux
Any browser
The board and the workflow
Workflow stages
Pending → Todo → Done
Six stages, incl. Review
Agent Kanban ships six: Backlog → Refinement → Ready → In Progress → In Review → Done. The Ready stage is the queue agents poll. PortBay's board is deliberately three columns because a column change is what triggers a run.
Priority levels
No
Low → Critical
Activity log of every change
Git history of the files
Yes — human and agent
Where tasks are stored
Markdown in your repo
agent-kanban.io servers
PortBay writes each card as a markdown file under .portbay/tasks/, so it is diffable and reviewable in a pull request. Agent Kanban stores tickets on its own servers, with markdown supported in descriptions and comments.
Stall detection
No
Nudges on stalled work
Agent Kanban advertises nudges when work stalls — useful when nobody is watching an unattended agent.
Agents & the API
MCP in both directions
Yes — agents read/write cards
Yes — 5 hosted MCP tools
Agent Kanban's hosted MCP server exposes list_ready_tickets, get_ticket, update_ticket, add_comment and create_ticket, connected with one `claude mcp add --transport http` line. PortBay's MCP server lets a dispatched agent read and update its own card.
Plain REST + OpenAPI spec
MCP + CLI
Yes, generated spec
Every Agent Kanban MCP tool is also a REST endpoint, with an auto-generated OpenAPI spec — so an agent with no MCP client at all can still work the board.
Purpose-built polling endpoint
Push, not poll
Yes — /tickets/ready
/tickets/ready exists so an unattended agent can ask "is there work?" cheaply on a schedule. PortBay has no equivalent because it pushes work to the agent instead of waiting to be asked.
Agents the tool itself dispatches
11 CLIs + any via Custom
None — by design
A different question from which clients can read the board. Agent Kanban names Claude Code, Cursor, Claude Desktop and Codex as clients, and ships a Claude Code skill, but launches none of them.
The environment the agent works in
Managed PHP / Node runtimes
Yes, pinned per project
No
Per-project database
7 engines, per project
No
PortBay provisions and supervises the engine — MySQL, MariaDB or PostgreSQL and four more — with an isolated data directory and injected connection variables.
Trusted local HTTPS + real domain
mkcert on yourapp.test
No
Captured mail + public tunnel
Both, one click
No
Parallel agents in git worktrees
Yes — worktree per card
Not its layer
PortBay puts each dispatched card in .portbay/worktrees/<card-id> on a card/<id> branch, under a board-level concurrency cap. Agent Kanban leaves isolation entirely to whatever is running the agent.
OS-level sandbox around the run
Seatbelt + network policy
Not its layer
PortBay generates a macOS Seatbelt profile per project: writes confined to the project folder, a network policy of blocked, loopback, outbound or full, denials logged line by line.
Price, licensing & ownership
Free tier
Community, 6 projects
1 project, 50 tickets/mo
Agent Kanban Free is $0: 1 project, 50 tickets per month, 1 agent API key, full API access, activity history and the Claude Code skill. PortBay's Community plan covers 6 projects with no ticket cap.
Paid price
$10/mo
$5/mo, unlimited
Checked 2026-09-06: Agent Kanban Pro is $5/month for unlimited projects, tickets and agent API keys. PortBay Pro is $10/month — twice the price, for a product doing a much larger job.
Free perpetual licence route
Merged PR or sponsor
Not offered
Open source
Yes — AGPL-3.0
No — hosted SaaS
You hold the data
Files in your repo
Export via REST API

Agent Kanban facts on this page were last checked against agent-kanban.io on . Pricing and limits change — check their site before you decide.

Migration

Already using Agent Kanban?

There is a good chance you should keep it. The two tools answer different questions, and the split is clean: Agent Kanban is the queue that stays reachable when your Mac is asleep — cron jobs, CI runs, tickets you file from your phone — and PortBay is the machine that runs the cards needing a live application underneath them. If you do want to consolidate, moving out is a read job rather than a migration, because every board operation is a REST endpoint with a published OpenAPI spec: an agent can list your open tickets and write the same cards into PortBay as markdown files in .portbay/tasks/, where they become part of the repository. Either way, nothing about your agent CLIs, API keys or git remotes changes.

Agent Kanban
1

Install PortBay and add the repository the tickets are actually about.

2

Press play — PortBay provisions the runtime, a database and trusted HTTPS for that project.

3

Have your agent, still connected to Agent Kanban over MCP, read the Ready queue and open matching PortBay cards; move one to Todo and let PortBay dispatch it against the running app.

PortBay
Common questions

PortBay vs Agent Kanban, in plain terms

Don’t see your question answered here? The docs go deeper — every feature, flag and failure mode.

Read the docs

No — and it is worth correcting, because the name suggests otherwise. Agent Kanban describes itself as "the hosted task queue for coding agents". Tickets live on agent-kanban.io, the MCP server is remote, and the product's argument is explicitly against the local model: it says tools that run agents on your laptop leave you with a board that "dies when the lid closes". PortBay is the local-first one here: cards are markdown files in your repository, the board runs in a desktop app on your Mac, and nothing needs a network to open. Both positions are defensible; they are just opposite.

Yes, for the thing they are describing. If your workflow is an unattended agent polling for work on a schedule from a server or a CI runner, a hosted queue is the right shape and a desktop board is the wrong one. PortBay's board is unreachable when the Mac is asleep, and no amount of framing changes that. What the local model buys is the other side: the agent runs inside an environment PortBay has already provisioned, so it can apply a migration, hit the page over HTTPS and read the captured mail — and the run itself is confined to a git worktree inside a macOS Seatbelt profile with a network policy. If your work needs a running application, having it locally beats having the board remotely.

No, and it does not claim to. It is the queue; the agent runs wherever you already run it and connects over MCP or plain REST. Its documented loop is the agent's: poll /tickets/ready, claim the ticket by setting itself as assignee and moving to in_progress, do the work, add a comment with the pull-request link, move to in_review. PortBay does the opposite — a card moved to Todo starts the agent itself. It dispatches 11 out of the box — Claude Code, OpenAI Codex, Cursor, Gemini, Aider, Copilot CLI, OpenCode, Amp, Qwen Code, Antigravity and Ollama — plus any other CLI via a Custom command template.

Agent Kanban, straightforwardly. Its free tier is $0 for 1 project, 50 tickets a month and 1 agent API key, and Pro is $5/month for unlimited projects, tickets and keys, with no credit card needed to start. PortBay is $10/month for Pro, free on the Community plan up to 6 projects, and free for life with a merged qualifying pull request or a GitHub sponsorship. The prices are not really comparable, though: one is a ticket database with an API, the other also installs and supervises runtimes, seven database engines, a certificate authority, a mail catcher and a tunnel. Pick on what you need, not on the five dollars.

Reachability, workflow granularity and API surface. It is a URL, so a cron job, a CI runner, a Linux box and a phone all see the same queue. Its workflow has six stages including Refinement and In Review, with Low/Medium/High/Critical priorities, an activity log covering both humans and agents, and nudges when work stalls — PortBay's board is three columns because a column change is what triggers a run. And every MCP tool is also a REST endpoint with an auto-generated OpenAPI spec, so an agent with no MCP client can still work the board. It also runs anywhere a browser does, including Windows, which PortBay does not.

Yes, and for a lot of setups it is the right answer. They sit at different layers and neither takes ownership of your git remote or your agent subscriptions. Keep Agent Kanban as the always-on queue that CI and scheduled jobs write to, and let PortBay handle the cards where the agent needs a database, a runtime and a live URL. An agent connected to both can read a Ready ticket and open the matching PortBay card with nothing custom in between.

Agent Kanban names those two itself, as the local orchestrators it is positioned against. Vibe Kanban and Conductor are the same shape as PortBay — desktop tools that run agents on your machine — so those comparisons turn on what surrounds the agent, and PortBay's answer is the provisioned environment. Agent Kanban is a different shape entirely: it never runs an agent, so this comparison turns on hosted-and-always-reachable versus local-and-actually-running. We publish both sets of pages, and if you are choosing between a hosted queue and a local runner, that is the first decision to make.

PortBay mascot — a friendly blue tugboat

Give your projects and your agents a real local home.

Download for macOS

Free & open source · macOS 11+ on Apple Silicon · Pro from $10/mo