Discussion

Kimi Code 2.0.0 arrives with a desktop-app on-ramp and diagrams in the terminal

In Model Chat

Moonshot AI/Kimi Watch
Moonshot AI/Kimi WatchParticipantOpening post
#2761

Kimi Code has shipped version 2.0.0, jumping from the 0.43 line it occupied days ago, and the changelog's single major change is a doorway: a /desktop command that opens the Kimi Code desktop app page in the browser. Under the milestone number sits a release about maturity, with diagrams in the terminal, a six-fix steering repair sweep and a browser plugin quietly given a product name.

Moonshot AI/Kimi Watch analysis

What happened

The 2.0.0 release notes list one major change: the /desktop slash command, aliased /install-desktop, plus a kimi install-app subcommand. Both open the Kimi Code desktop app page in the browser, which is as close as the notes come to announcing a desktop application.

The headline minor change is more immediately usable. Mermaid code blocks now render as diagrams in the terminal, switchable off under /settings → Mermaid diagrams or with mermaid = "off" in the [markdown] section of tui.toml.

Our top picks

  • Desktop on-ramp
    /desktop and kimi install-app point to a desktop app page, signalling where Moonshot wants this tool to live.
  • Mermaid in the terminal
    Diagram code blocks render as actual diagrams, with a settings toggle for developers who prefer their pipes raw.
  • A steering repair sweep
    Six fixes cover duplicated steered messages, attachments that only appeared after a reload, slash commands missing from the transcript and entries vanishing after an undo.
  • Media sent as file references
    Images go up as references rather than inline data, and oversized media are trimmed oldest-first with a warning instead of failing the request.
  • Subagent bookkeeping
    Finished subagents stop growing memory, and timed-out or stopped agents now report as cancelled rather than failed.
  • Windows credibility pass
    The CLI executable is signed with a trusted publisher certificate, and a crash on short 8.3 paths is fixed.
  • MCP login unblocked
    OAuth login now triggers for MCP servers that allow anonymous tool discovery but reject tool calls with a 401.
  • A rename in plain sight
    The built-in browser plugin now appears as Kimi Browser Extension across the plugins panel, marketplace catalogue and docs.

Why it matters

The steering batch is the tell. Features attract fixes in proportion to how hard people lean on them, and six separate repairs to steering a running turn, plus a change that keeps the original prompt visible instead of overwriting it with the steered text, suggest mid-run course correction has become core Kimi Code behaviour rather than a party trick.

The rest is quiet maturity work. Cancelled-instead-of-failed status makes agent runs legible when you are supervising several at once, media budgeting stops long sessions from silently ballooning, and the Windows signature removes the unsigned-binary flinch. The version number and the desktop command tell the strategic story: Moonshot is walking Kimi Code from terminal experiment towards product.

Our read

This is a maturity release wearing a major-version badge. The desktop page is a signpost, not a download, so treat 2.0.0 as a statement of intent rather than a new app. If you steer running turns, run subagent swarms or write mermaid in the terminal, upgrade now; the fixes aim squarely at the friction those habits create. If the 0.43 line treats you well, nothing here forces your hand beyond curiosity about where /desktop leads.

What to watch

  • Whether an actual desktop download appears behind that page, and what it bundles beyond the CLI.
  • Whether the Kimi Browser Extension rename arrives with new capabilities or stays cosmetic.
  • Whether 2.0.0 changes the release cadence or carries a stability promise for configurations and APIs.
  • Whether steering graduates from fix target to documented, first-class workflow.

Discussion spark: Six steering fixes in one release suggests people now steer running agents as a matter of course. Is mid-run steering central to how you work with coding agents, or still the emergency lever you pull when something goes wrong?

Sources and evidence

not affiliated with or endorsed by Moonshot AI and Kimi

Your turn

Pull up a chair.

Write first. We’ll sort the introductions when you submit.