What shipped in the desktop app, newest first. Each release keeps a permanent page of its own, and the feed carries the whole note rather than a teaser.
Pekan.cloud now has documentation. Until today the site could tell you what Pekan was and what had shipped in it, but nothing on it explained how to actually do anything: how to install it, how to wire a coding agent to a board, or what any of the views were for. Seven pages at pekan.cloud/docs.html close that gap.
Sakura joins Light, Ember, Midnight, and Sepia as a built-in theme, and it is light only. It takes its cue from cherry blossom rather than from a tinted grey: page, window chrome, panels, columns, and cards sit on soft blossom neutrals, hover and selection surfaces warm gently toward rose, and text is a deep warm plum that stays crisp on every one of those surfaces. The accent is a muted rose gold, restrained enough to read as chrome rather than decoration, so buttons, the active navigation item, focus rings, and the terminal cursor all stay recognisable without turning the workspace candy pink.
Custom agent actions no longer have to pretend that every job ends as a task update, a new task, or a project-context change. New actions now default to Not specified (follow prompt), which adds no result-target instruction and lets the action prompt define the work. Project actions can also explicitly target Repository changes when the intended result is code, configuration, a version bump, or Git work. The distinction between result and permission remains visible: Propose produces a reviewable answer without changing the board, files, or Git, while Apply permits only the changes the prompt requests. The action editor now explains Context, Result target, and Behavior directly beneath each choice, and launch confirmation calls out repository and open-ended permissions before a run starts.
The Copy link to task action in a task's comment menu used to hand you a URL built from whatever address the app happened to be served from. In the desktop app that address is a loopback port on your own machine, so the link you copied looked like http://localhost:3847/tasks/… and opened nothing at all for the person you sent it to. The action promised sharing and quietly delivered a dead link.
This release takes two surfaces out of Pekan and folds a third into where it belonged all along. The project Queue is gone. It was a second, hand-maintained ordering of the same tasks that already sit on your board, and because it never claimed, assigned, started, or completed anything, its sequence could quietly drift away from reality without changing a single outcome. Ordering work now happens where it always actually happened: board columns, task priority, dependency edges, and Roadmap milestones. Every task, description, assignment, priority, dependency, milestone link, and activity record is untouched. Only the queue membership and its order were removed, and the migration that drops those tables cannot touch a task, because nothing outside them ever pointed in. Opening an old /queue link lands on the project Tasks view instead of a dead page.
The built-in Dark theme has been retired. It had drifted to within a few percent of Ember (same warm charcoal canvas, same amber-orange accent, differing mainly in how dark the window frame sits), and two themes that close apart are a choice nobody can make meaningfully. Ember stays: near-black chrome across the title bar, tab strip, sidebar, and status bar, a warm charcoal workspace, and the brighter ember accent. Light, Midnight, and Sepia are unchanged, so the picker in Settings → Appearance now offers four built-in themes rather than five.
Pekan can now show an agent's live intent before its work produces a comment, assignment, or card move. Agents connected through MCP can announce that they are viewing a task, planning an approach, or editing work. That short-lived state appears as an amber pulse on the task card and in the task-detail header, with the agent name and any concise context available on hover. It is deliberately separate from the green Agent working badge: the existing badge describes a persistent task claim, while presence describes what a live agent session says it is doing right now.
Pekan can now capture what a terminal Claude Code session actually cost, without anyone having to estimate anything. When a session ends, a hook reads the transcript Claude Code just wrote, totals the token usage recorded in it, and reports that to your board. The numbers come from the provider's own accounting, so the Telemetry view's Runs tab fills in for sessions the desktop app never launched. For most people, that is nearly all of them.
Pekan has a new workspace tab, Telemetry, sitting between Roadmap and Timeline. It answers the questions a kanban board is uniquely placed to answer and almost never does: where work actually piles up, how long it really takes, how often it comes back from review, and how far token estimates land from reality.
Agent run output (the log block that opens under a run on the Activity page and in the Agent runs panel) was painted a fixed near-black with pale blue text, left over from the interface Pekan used before the redesign. It ignored the theme completely, so it stayed the same cold slab whichever theme you picked, and read especially oddly in the light themes. It now uses the theme's own colours and sits recessed inside the card that contains it, so it reads as an inset well of machine output rather than a foreign block. The two copies of it, which had drifted into slightly different styling, are now one component.
The app icon and the Pekan mark still carried the blue-violet accent the interface moved off in 0.158.0, which left a purple P sitting next to an amber workspace on the taskbar, in the title bar, and on the website tab. The mark is now warm throughout: a warm near-black plate, a warm off-white P, and the amber brand in the bowl with a lighter amber fold catching the light. The shapes are untouched: this is a recolour, not a redraw, so the mark stays recognisable.
The right-hand Files sidebar was the one part of the workspace the new design did not reach, and it showed: three stacked header rows (the project name, the Files and Diff switcher, and a row of file actions) where the rest of the window had already collapsed to one. It now matches. A single row carries the Files and Diff switcher on the left, with the changed-file count riding alongside the Diff label, and the new file, new folder, quick open, and refresh actions on the right, at exactly the height of the view header across the pane. The project name moves to its own line below in monospace, alongside the control that collapses the sidebar.
Pekan has been redrawn around a warmer, more deliberate visual language. The interface moves off its blue-violet accent onto a warm neutral base with an amber accent, and the typeface changes from Geist to IBM Plex Sans, with IBM Plex Mono carrying every piece of machine data: task identifiers, dates, estimates, token counts, column counts, and branch state. That split is the point. Prose reads as prose, and anything a machine produced reads as machine output, so you can scan a column for identifiers or costs without reading a single word of a title.
Agent setup for remotely reachable projects now writes the public board address into .mcpekan/config.json. Previously a remote server advertised whatever address it had bound locally, which behind a reverse proxy is a loopback upstream such as http://127.0.0.1:3847/mcp. Repair Agent Setup copied that value verbatim into the project, so a correctly issued access token was paired with an address pointing at the developer's own machine. Coding agents then either failed to connect or reached an unrelated local MCPekan that rejected the token, producing an Unauthorized error that gave no hint the URL was the problem. Remote deployments now declare their public MCP URL explicitly and refuse to start without it, so the failure surfaces once at boot with a clear message instead of silently spreading into every project configuration.
Agent-assisted project planning now pauses when the draft raises consequential questions instead of showing them as static text at the end. You can answer each question, ask the selected agent to revise the plan around those decisions, and review the resulting goals, milestones, and tasks before applying anything. If you intentionally want to proceed with unresolved questions, that remains an explicit option. Closing or discarding at this stage cleans up the pending proposal without changing the project.
The roadmap now speaks in plain language: a goal is an outcome you want to reach, milestones are the concrete checkpoints toward it, and board tasks link to a milestone to track the real work. "Goal" simply replaces the old "initiative" wording everywhere you see it: your existing roadmap data is untouched and nothing needs migrating. You can create and maintain goals and milestones by hand from the Milestones panel with a new "+ New goal" form, no agent required, and the roadmap's empty state now explains the goal → milestone → task model and offers both a manual start and an agent-assisted one.
Delivered agent work that needs a human decision now records a lasting review trail instead of living only in memory. When an isolated run is pushed for review, Pekan writes a review artifact keyed to the task, run, and the immutable base commit it was cut from, and it stamps a plain-language event onto the task timeline, "Delivered work for review", alongside the diff statistics (files changed, insertions, deletions). Re-delivering the same run against the same base resolves to the one existing review rather than piling up duplicates, so the queue stays honest across retries and restarts. Auto-merged clean runs still bypass the queue entirely; only work governed by a human-review policy or explicitly escalated creates a review.
Pekan now labels AI-authored work in plain language wherever you read it. Tasks created by an AI agent carry an AI-generated badge on the board, in the task list, and on the task detail page. Agent-written handoffs use the same explicit language, and the Agent Actions library now explains that running an action produces AI-generated output. The distinction is intentional: operational badges can still describe an agent as working or stale, while content labels tell you who created what.
Pekan now gives agent-driven work one deliberate place to ask for human attention. The new Needs Human feed gathers the conditions that should interrupt you without mixing them into ordinary notifications: delivered changes waiting for review, failed or timed-out agent runs, task claims whose agent has gone quiet, and structured questions that need an answer. A badge in the app header shows the current open count, and selecting an item takes you directly to the relevant review or task. The feed is global across your visible projects, so you no longer have to open each board and scan for something an agent left behind.
Pekan now has a proper account section in Settings. It shows who you're signed in as and which plan you're on, in one place, at a glance. Signing in opens your browser, you authenticate on the web the way you would with any other site, and Pekan picks the session back up automatically when you return. There's no password typed into the desktop app, and there's no license key to paste anywhere: your access follows your account.