VibePager · Agents

Running Kimi CLI from your phone

Kimi is the open-weight agent that keeps showing up at the top of coding benchmarks at a fraction of the price — which is why people run K3 for long jobs, and why they need to be told when it stops.

Updated September 2026

What Kimi CLI is

Moonshot AI’s terminal coding agent, running the Kimi family of open-weight models. It comes as a standalone CLI and also inside its editor extension; VibePager detects both. It speaks the Agent Client Protocol, so it is driven over a real protocol.

The reason people reach for it is price-per-capability: the Kimi models — K3 as of September 2026, K2.5 and K2 Thinking before it — have repeatedly landed near the top of open-weight coding benchmarks while costing far less than the closed frontier models, and the weights are published, so you can also self-host.

Kimi CLI vs Claude Code

Setting it up

1. Have Kimi CLI installed and signed in on your Mac or PC, the way you already run it.
2. Install VibePager on the same machine. It finds the agents that are there and lists only those.
3. Open vibepager.com/app on your phone, pair once with the code shown on the computer, pick a folder, and start.

A note on model versions

Kimi ships new model versions fast, and so do GLM, DeepSeek, Qwen and MiniMax. VibePager deliberately does not hardcode a model list: the phone shows what your installation offers today, with search. See the models page.

What is the same for every agent

• Kimi CLI runs on your machine, on your files, with your own account or API key. Nothing is uploaded.
• Phone-to-machine traffic is end-to-end encrypted; the relay in the middle has no key.
• You choose which folders the phone is allowed to open.
• The app lists only the agents actually installed — an agent you do not have never shows up as a broken option.
• One subscription covers up to 3 computers and 6 phones or browsers.

Questions

Can I control Kimi CLI remotely?

Yes — it runs on your machine and the phone sends tasks and answers its permission prompts, over the Agent Client Protocol.

Kimi CLI vs Claude Code?

Kimi runs open-weight models at a much lower price and can be self-hosted; Claude Code is closed, more expensive, and generally stronger on the hardest tasks. Many people install both.

Is Kimi free?

The weights are open, so self-hosting is free apart from the hardware. Moonshot’s hosted API is paid, usually with a free allowance to start. VibePager uses whatever sign-in you already have.

Can I use Kimi models inside Claude Code instead?

That depends on Claude Code’s own configuration, not on VibePager. What VibePager guarantees is that whichever agent you run, the phone shows the models that agent actually offers on your machine.