Moonshot AI's Kimi Code Now Lets You Control Coding Sessions From Any Device

Moonshot AI just shipped one of the most requested features in its coding CLI. Kimi Code 0.39.0 introduces Remote Control, an experimental mode that lets you keep a session running on your workstation while you drive it from a laptop on the couch, a tablet in a meeting, or a phone on the train.

The capability has sat on the community wishlist for months. A GitHub issue filed earlier this year framed it bluntly: Claude Code implements this as remote control with bridge-backed sessions that allow multi-device attach and mobile push notifications, and that UX sets a strong baseline for what developers expect from an agent CLI in 2026. Kimi's answer arrives with 0.39.0.

Flipping the switch

Remote Control is gated behind an environment variable while it stabilizes. Set KIMI_CODE_EXPERIMENTAL_REMOTE_CONTROL=1, then run kimi rc, kimi web --remote-control, or /remote-control to access a local web session remotely.

In practice that looks like:

export KIMI_CODE_EXPERIMENTAL_REMOTE_CONTROL=1
kimi rc
# or, from an existing web session:
/remote-control

Once enabled, the local session becomes reachable from any browser you point at it, and the same session state stays synchronized across every client that connects.

What got fixed under the hood

Two smaller entries in the changelog are worth pulling out because they make the feature work end to end. The first is a fix for messages sent from one web client not appearing on other clients connected to the same session, which is the core multi-device sync problem you would hit the moment you opened the same session on a phone and laptop. The second is a security tightening: the --allow-remote-terminals flag was removed from kimi web, and PTY terminal routes now stay available on loopback binds only. Arbitrary remote shells are off the table, and the intended path is Remote Control itself.

The problem it solves

Terminal-native agents have an ergonomic ceiling. Your session is pinned to the machine that spawned it, so if you close the lid or move rooms, you either lose context or leave a laptop tethered somewhere. The community request that seeded this feature laid out the pain points clearly, from stepping away while the agent is working and being unable to approve critical tool calls from another machine, to switching devices mid-session, to monitoring background work without keeping a terminal window pinned to one machine. Remote Control turns the local Kimi process into something closer to an ambient service you attach to.

The rest of 0.39.0 packs a punch

Remote Control headlines the release, though a few sibling features shipped alongside it that agent builders will care about:

• Tower mode for multi-agent orchestration. Set KIMI_CODE_EXPERIMENTAL_TOWER=1, then use /tower on and /tower to start Tower mode and coordinate multiple agents toward an objective.
• Subagent fork. Set KIMI_CODE_EXPERIMENTAL_SUBAGENT_FORK=1, or subagent_fork = true under [experimental], to let the subagent and swarm tools use fork to start from a snapshot of the calling agent's conversation history. This is the CLI equivalent of cloning a git branch mid-thought, so a subagent can explore a tangent without polluting the parent's context.
• Independent swarm timeouts. Configure an independent AgentSwarm timeout with [swarm] timeout_ms or KIMI_CODE_SWARM_TIMEOUT_MS.
• Windows path fixes. File tools and shell now resolve Git Bash paths like /c/Users or /tmp correctly.

Sharp edges to watch

This is an experimental feature, and the flag name is honest about that. A few things worth keeping in mind:

1. It exposes a local web session over the network, so treat the URL like an SSH key. Removing --allow-remote-terminals narrows the blast radius, though Remote Control still gives whoever holds the endpoint the ability to steer your agent.
2. Feature parity with the terminal is the goal, though the mobile web has only just gotten its polish pass in this release, including grab-handle bottom sheets for the slash-command and mention menus and a revamped right sidebar.
3. The variable name suggests behavior may change or move behind different flags in future releases, so don't bake it into CI yet.

For anyone already using Kimi Code as their primary agent CLI, this update finally decouples your session from the box under your desk. A small feature flag with a large workflow implication, it closes one of the last obvious gaps between Kimi Code and its closest commercial competitor.