Cadre

Cadre is a desktop app that puts a Kanban board in front of your coding agents. Instead of juggling separate terminal sessions for Claude Code, Codex, and whatever else you run, you write tasks as cards, route each card to an agent, and watch the work move across columns while you handle the parts that actually need a human, answering questions, reviewing plans, and approving diffs. It's project management software where the workers happen to be AI agents.

The problem it goes after is real for anyone who has leaned into agentic coding. One agent in one terminal is manageable. Three agents across two repos is a mess of windows where you lose track of what's waiting on you, what's running, and what quietly stalled an hour ago. Cadre turns that sprawl into a board with custom columns, sections, and tags, so the state of every task is visible at a glance and parallel runs stop being something you track in your head. The mess compounds because agents are slow in an awkward way, fast enough that you can't walk away, slow enough that you tab elsewhere and forget them. A board that surfaces exactly what stalled and why is the difference between actually running three agents and merely owning three agents.

The Kanban framing sounds cosmetic until you use agents at any volume. Agent work has natural states, queued, planning, waiting on approval, executing, blocked on a question, done, and those states map cleanly onto columns. Sections and tags let you slice the board by project or priority, and routing decides which agent picks up which kind of card. The question you ask all day changes from what is that terminal doing to what needs me right now, which is the right question once the agents outnumber you.

Human oversight is structured rather than assumed. There are four kinds of approval gates where a task pauses for you. An agent can ask a question, present a plan for sign-off, request permission to run a command, or hand over a diff for review before anything proceeds. That's the piece that makes parallel agents tolerable, since you're not trusting five autonomous processes to behave, you're reviewing checkpoints that come to you on the board. In practice you spend your attention where the risk lives, waving a routine chore through its checkpoints in seconds while a risky migration gets a hard look at every gate.

On the agent side it's pointedly neutral. Cadre routes work to Claude Code, OpenAI's Codex, Grok, Opencode, and any OpenAI-compatible endpoint, running them through your existing CLI logins or saved API credentials. You bring your own keys, tokens are billed by your provider, and Cadre is explicit that it charges for the orchestration layer, the boards, parallel runs, and approvals, never per token. Credentials are stored in your OS keyring, and the whole app is local-first and works offline by default, with your repositories, worktrees, and credentials never leaving your machine. That neutrality is worth something on its own. Agent tooling is churning fast, and a board that treats agents as swappable adapters means switching from one CLI to another doesn't mean rebuilding your workflow.

The audience is developers who have moved past trying a coding agent and into depending on several. Solo developers get a control tower for parallel work across projects. Teams get the optional online workspaces, which add cross-device sync, workspace invites, and shared board activity, with only the workspace data you choose ever being synced. Compared to terminal multiplexers, homegrown scripts, or cloud agent platforms that want your repo on their servers, Cadre's pitch is a familiar project-management surface bolted onto local execution, which is a combination not much else offers right now.

Access is friendly to trial. The local desktop version is free, with local boards, local agent adapters, and the full set of approval gates, and no credit card is needed to start. Pro and Team tiers add the online workspaces and collaboration features, though the site doesn't publish their prices. It ships for Windows x64, macOS as a universal binary, and Linux as both AppImage and .deb, so whatever machine your agents live on, the board can live there too. Nothing about the free tier reads like a demo either, it's the actual product, with the paid tiers selling sync and collaboration rather than unlocking the core. For a young product in a fast-moving category, starting free and local is exactly the right posture, since you can adopt it without handing anything over.