Kimi Code Setup: API Key, K3.1 Pro, and Claude Code

To use an I Code Easy API key in Kimi Code CLI, add the key and a URL ending in /v1 to ~/.kimi-code/config.toml, then select k3.1-pro. In Claude Code, use ANTHROPIC_* variables and the domain alone. The two clients use different configuration formats.

Choose your client and setup path

Register, copy your site key from API Keys, and select an endpoint for Kimi in Endpoint selection. An I Code Easy API key and an official Kimi account are separate credentials. /login signs in to the official account; it is not where you enter this site’s key.

1. Install Kimi Code

macOS / Linux:

curl -fsSL https://code.kimi.com/kimi-code/install.sh | bash

Windows PowerShell:

irm https://code.kimi.com/kimi-code/install.ps1 | iex

Kimi Code on Windows also requires Git for Windows; check the official installation instructions for requirements.

Open a new terminal after installation, then run the version command. If kimi is not found, resolve installation or PATH before changing API settings.

kimi --version

Check the official Kimi Code page for installation and update instructions. If your version rejects these settings, compare its documentation and installed version before repeatedly replacing the key.

2. Configure the I Code Easy API

Create ~/.kimi-code/config.toml if needed; on Windows this is .kimi-code/config.toml under your current user’s home directory. Merge into an existing file, preserving other providers, MCP settings, and permission rules. Keep one top-level default_model before any TOML tables, and do not append duplicate provider or model tables.

If KIMI_CODE_HOME is set, it changes the configuration directory. default_model takes the model table alias, such as icodeeasy/k3.1-pro, rather than the wire model name alone.

default_model = "icodeeasy/k3.1-pro"

[providers.icodeeasy]
type = "kimi"
base_url = "https://api.icodeeasy.cc/v1"
api_key = "Your I Code Easy API key"

[models."icodeeasy/k3.1-pro"]
provider = "icodeeasy"
model = "k3.1-pro"
max_context_size = 1048576
capabilities = ["thinking", "always_thinking", "image_in", "video_in", "tool_use"]
display_name = "Kimi K3.1 Pro (I Code Easy)"
support_efforts = ["low", "high", "max"]
default_effort = "high"

[models."icodeeasy/k3.1-pro-highspeed"]
provider = "icodeeasy"
model = "k3.1-pro-highspeed"
max_context_size = 1048576
capabilities = ["thinking", "always_thinking", "image_in", "video_in", "tool_use"]
display_name = "Kimi K3.1 Pro Highspeed (I Code Easy)"
support_efforts = ["low", "high", "max"]
default_effort = "high"

[models."icodeeasy/k3"]
provider = "icodeeasy"
model = "k3"
max_context_size = 1048576
capabilities = ["thinking", "tool_use"]
display_name = "Kimi K3 (I Code Easy)"

[models."icodeeasy/k3-256k"]
provider = "icodeeasy"
model = "k3-256k"
max_context_size = 262144
capabilities = ["thinking", "tool_use"]
display_name = "Kimi K3 256K (I Code Easy)"

On macOS / Linux, restrict file permissions:

chmod 600 ~/.kimi-code/config.toml

If the primary domain is slow, use https://jp.icodeeasy.cc/v1 or https://sg.icodeeasy.cc/v1 as base_url. Do not append /v1 twice.

Choose among the four models

See pricing and usage for current prices, balance, and usage, and the model catalog for supported capabilities. Image and video input are not publicly promised for K3 or K3-256K.

Legacy names do not all select the same model. Legacy K2 names, kimi, and kimi-for-coding route to K3.1 Pro; Highspeed aliases such as kimi-for-coding-highspeed select Highspeed. kimi-k3 and k3[1m] remain K3, while kimi-k3-256k remains K3-256K. The corresponding kimi--prefixed names for all four tiers retain their model selection.

When K3-256K exceeds its context limit, the existing mechanism upgrades the request to K3 and bills that request at K3 prices. Start a new session after switching models to avoid carrying over previously compressed context.

Temporary test configuration

Current Kimi Code does not automatically read ordinary KIMI_API_KEY. Use the dedicated KIMI_MODEL_* variables for a temporary trial. This is a separate setup path from the config file; identify which path you are testing first.

Setting KIMI_MODEL_NAME activates the environment model and requires KIMI_MODEL_API_KEY. These variables do not write back to the file and take priority over its default_model; the command-line -m option takes priority over the environment model.

macOS / Linux:

export KIMI_MODEL_NAME="k3.1-pro"
export KIMI_MODEL_API_KEY="Your I Code Easy API key"
export KIMI_MODEL_PROVIDER_TYPE="kimi"
export KIMI_MODEL_BASE_URL="https://api.icodeeasy.cc/v1"
export KIMI_MODEL_MAX_CONTEXT_SIZE="1048576"
export KIMI_MODEL_CAPABILITIES="thinking,always_thinking,image_in,video_in,tool_use"
kimi

Windows PowerShell (current session only):

$env:KIMI_MODEL_NAME="k3.1-pro"
$env:KIMI_MODEL_API_KEY="Your I Code Easy API key"
$env:KIMI_MODEL_PROVIDER_TYPE="kimi"
$env:KIMI_MODEL_BASE_URL="https://api.icodeeasy.cc/v1"
$env:KIMI_MODEL_MAX_CONTEXT_SIZE="1048576"
$env:KIMI_MODEL_CAPABILITIES="thinking,always_thinking,image_in,video_in,tool_use"
kimi

For a temporary Highspeed trial, change KIMI_MODEL_NAME to k3.1-pro-highspeed. Closing this terminal removes the variables set here. Any values previously saved in user-level settings or startup scripts must be removed separately. Test TOML settings in a new terminal without these temporary variables.

3. Start and verify

Enter your project directory and run:

cd your-project-directory
kimi

Run /model in the interface and confirm Kimi K3.1 Pro (I Code Easy) or your intended tier. A version string or model list alone does not prove a successful API call.

Send a simple message such as “Reply only: connection successful” before asking the client to read or edit your project. After a reply, check status and actual model in request logs, then check usage details for usage and charges. Hide the key in any troubleshooting screenshots.

Use Kimi K3.1 Pro through Claude Code

For general installation and configuration, see the Claude Code guide. On macOS / Linux, add these Kimi settings to the ~/.bashrc or ~/.zshrc that your shell uses:

export ANTHROPIC_AUTH_TOKEN="your I Code Easy API key"
export ANTHROPIC_BASE_URL="https://api.icodeeasy.cc"
export ANTHROPIC_MODEL="k3.1-pro"
export ANTHROPIC_SMALL_FAST_MODEL="k3.1-pro"
export ANTHROPIC_DEFAULT_SONNET_MODEL="k3.1-pro"
export ANTHROPIC_DEFAULT_OPUS_MODEL="k3.1-pro"
export ANTHROPIC_DEFAULT_HAIKU_MODEL="k3.1-pro"
export CLAUDE_CODE_DISABLE_NONESSENTIAL_TRAFFIC="1"

Save, run source ~/.bashrc or source ~/.zshrc, and start claude. ANTHROPIC_BASE_URL uses the domain alone; alternatives are https://jp.icodeeasy.cc and https://sg.icodeeasy.cc. Start a new session after changing models.

On Windows, set the same variables as $env:VARIABLE="value" in PowerShell and launch claude from that window. If you save user-level permanent variables, open a new PowerShell window; running clients do not automatically receive them. Do not paste Bash export commands unchanged into PowerShell.

$env:ANTHROPIC_AUTH_TOKEN="your I Code Easy API key"
$env:ANTHROPIC_BASE_URL="https://api.icodeeasy.cc"
$env:ANTHROPIC_MODEL="k3.1-pro"
$env:ANTHROPIC_SMALL_FAST_MODEL="k3.1-pro"
$env:ANTHROPIC_DEFAULT_SONNET_MODEL="k3.1-pro"
$env:ANTHROPIC_DEFAULT_OPUS_MODEL="k3.1-pro"
$env:ANTHROPIC_DEFAULT_HAIKU_MODEL="k3.1-pro"
$env:CLAUDE_CODE_DISABLE_NONESSENTIAL_TRAFFIC="1"
claude

Troubleshoot 401, model selection, and URL errors

Configure another client

Claude Code · GLM setup · Pricing, balance, and usage · Model capabilities

Official references

• Kimi Code: Configuration files
• Kimi Code: Environment variables