Kimi Code CLI

Kimi Code CLI can use the Kimi platform’s pay-per-token API through the proxy’s moonshot path. The CLI has one credential slot, and here it holds your Milgram API key: you store your Kimi platform key in the console once, and the proxy puts it in place of your Milgram key on the way out. Milgram keeps it for your traffic only and never shows it again.

1. In the console, open your API key under Agents, choose Connect, and open the Kimi Code CLI tab.
2. On the Platform card, choose Use this, and store your Kimi platform API key (from platform.kimi.ai).
3. Configure the CLI as below.

Persistent

Merge this provider and model into ~/.kimi-code/config.toml and replace its top-level default_model. Keep every other setting in the file.

~/.kimi-code/config.toml

default_model = "milgram-moonshot"

[providers.milgram-moonshot]
type = "openai"
base_url = "https://api.milgram.dev/v1/proxy/moonshot/v1"
api_key = "mlg_YOUR_KEY"

[models.milgram-moonshot]
provider = "milgram-moonshot"
model = "kimi-k3.1-pro"
max_context_size = 1000000

One shell

The same provider and model for one shell, without changing config.toml: the CLI builds them in memory from these variables.

export KIMI_MODEL_PROVIDER_TYPE=openai
export KIMI_MODEL_BASE_URL=https://api.milgram.dev/v1/proxy/moonshot/v1
export KIMI_MODEL_API_KEY=mlg_YOUR_KEY
export KIMI_MODEL_NAME=kimi-k3.1-pro
export KIMI_MODEL_MAX_CONTEXT_SIZE=1000000

kimi-k3.1-pro is the model Kimi recommends today, with a 1M-token context; any model of the Kimi platform works the same way.

Note

The console’s Kimi Code CLI tab also offers a setup for Kimi Code subscriptions (signed in with kimi login). This guide covers the pay-per-token platform only.

Check it

Start kimi and send a prompt. The session appears in the console under Sessions, with the moonshot provider.