Kilo Code CLI

Overview

Kilo Code CLI is the terminal version of Kilo Code. It uses the same agent runtime and kilo.jsonc configuration as the VS Code extension, while keeping installation, model selection, task execution, and troubleshooting in the command line. This tutorial shows you how to:

Install and start the kilo command.
Connect EvoLink through an OpenAI-compatible provider.
Configure tool calling and token limits for Claude models.
Verify file tools from the terminal.

Prefer VS Code?

Follow the dedicated Kilo Code VS Code tutorial instead.

Before You Begin

1. Install Kilo Code CLI

npm
Install script
Homebrew

npm install -g @kilocode/cli

curl -fsSL https://kilo.ai/cli/install | bash

brew install Kilo-Org/tap/kilo

Verify the installation:

kilo --version
kilo --help

This guide targets Kilo CLI 1.0 and later. Run kilo upgrade if you have an older version.

2. Get an EvoLink API Key

Sign in to the EvoLink console.
Open API Keys and create a key.
Copy the raw key and keep it secure.

Store the key in an environment variable so it is not committed with your configuration:

macOS / Linux
Windows PowerShell

export EVOLINK_API_KEY="your_EvoLink_API_key"

$env:EVOLINK_API_KEY="your_EvoLink_API_key"

Set the raw key only. Do not add a Bearer prefix. Kilo’s OpenAI-compatible provider creates the authorization header automatically.

Step 1: Create the Kilo Configuration

Kilo CLI reads configuration from:

Global: ~/.config/kilo/kilo.jsonc
Project root: ./kilo.jsonc
Project directory: ./.kilo/kilo.jsonc

Kilo Code resolves {env:EVOLINK_API_KEY}-style references only in trusted config — the global config (~/.config/kilo/), a config passed via KILO_CONFIG, or organization-managed config. Project-level kilo.jsonc / .kilo/kilo.jsonc files cannot resolve {env:} references (this prevents a repository’s config from stealing credentials). Since this tutorial reads the key from an environment variable, save the following configuration to the global kilo.jsonc; project files are only suitable for overrides that don’t reference secrets, such as the default model:

{
  "$schema": "https://app.kilo.ai/config.json",
  "model": "evolink/claude-sonnet-5",
  "provider": {
    "evolink": {
      "npm": "@ai-sdk/openai-compatible",
      "options": {
        "apiKey": "{env:EVOLINK_API_KEY}",
        "baseURL": "https://direct.evolink.ai/v1"
      },
      "models": {
        "claude-fable-5": {
          "name": "Claude Fable 5",
          "tool_call": true,
          "reasoning": true,
          "limit": { "context": 1000000, "output": 32000 }
        },
        "claude-sonnet-5": {
          "name": "Claude Sonnet 5",
          "tool_call": true,
          "reasoning": true,
          "limit": { "context": 1000000, "output": 32000 }
        },
        "claude-haiku-4-5-20251001": {
          "name": "Claude Haiku 4.5",
          "tool_call": true,
          "reasoning": true,
          "limit": { "context": 200000, "output": 32000 }
        },
        "claude-opus-4-8": {
          "name": "Claude Opus 4.8",
          "tool_call": true,
          "reasoning": true,
          "limit": { "context": 1000000, "output": 32000 }
        }
      }
    }
  }
}

Key details:

Field	Purpose
`npm`	Uses the OpenAI Chat Completions-compatible protocol
`options.apiKey`	Reads the key from `EVOLINK_API_KEY`
`options.baseURL`	Uses the EvoLink API root ending at `/v1`
`model`	Sets the default using `providerID/modelID` format
`tool_call`	Enables file, edit, terminal, and other agent tools
`limit.context`	Enables context tracking and compaction
`limit.output`	Sets the response budget; Kilo defaults to a 32K cap

Use https://direct.evolink.ai/v1 as the Base URL. Do not append /chat/completions. Model settings must remain under provider.evolink.models.

Step 2: Validate the Configuration

Check the JSONC structure:

kilo config check

Inspect the models Kilo loaded for EvoLink:

kilo models evolink --verbose

You should see models such as evolink/claude-sonnet-5 with their configured context and output limits. After changing configuration, restart Kilo or run /reload from an active interactive session.

Step 3: Start the Terminal Agent

Open the project you want Kilo to work on:

cd /path/to/your/project
kilo

Run /models and confirm that evolink/claude-sonnet-5 is selected. Then send a read-only task:

Read package.json in the project root and tell me its name field. Do not modify any files.

Successful configuration means:

Kilo invokes a file-reading tool instead of returning only conversational text.
The value matches the project file.
No 401, 404, model_not_found, or unsupported-tool error appears.

Step 4: Common CLI Workflows

Start the interactive terminal UI:

kilo

Useful slash commands:

Command	Purpose
`/models`	Switch models
`/agents`	Switch Code, Plan, Debug, and other agents
`/status`	Inspect the current session
`/review`	Review local code changes
`/compact`	Compact a long session
`/reload`	Reload configuration, skills, and agents
`/exit`	Exit Kilo

Run a one-off task without entering the full TUI:

kilo run -m evolink/claude-sonnet-5 "Find the cause of the failing tests without changing files"

Temporarily select another model:

kilo -m evolink/claude-fable-5
kilo run -m evolink/claude-haiku-4-5-20251001 "Summarize this repository"

Suggested model roles:

Agent or task	Recommended model
Planning and complex reasoning	`evolink/claude-fable-5`
Everyday implementation	`evolink/claude-sonnet-5`
Lightweight exploration	`evolink/claude-haiku-4-5-20251001`
High-difficulty coding	`evolink/claude-opus-4-8`

Troubleshooting

`kilo: command not found`

Make sure the global npm executable directory is on PATH, then restart the terminal. Use npm prefix -g to locate the global installation directory.

`kilo config check` reports errors

Check JSONC commas and braces, confirm that provider, models, and options have the same nesting as the example, and use evolink/model-id for the default model.

`401 unauthorized`

Confirm that EVOLINK_API_KEY exists in the same terminal session that starts Kilo. Do not print the full key into logs or screenshots, and do not add a Bearer prefix.

`404` or a duplicated request path

Set baseURL back to:

https://direct.evolink.ai/v1

Do not use /v1/chat/completions as the Base URL.

`model_not_found`

Model IDs must exactly match EvoLink. Run kilo models evolink --verbose and do not add an anthropic/ prefix or remove date suffixes.

Chat works but file tools do not

Confirm that the model includes "tool_call": true. Retry with claude-sonnet-5 or claude-fable-5 if a weaker model cannot reliably drive tools.

Long sessions exceed the context window

Confirm that provider.evolink.models.<modelID>.limit.context exists and is not 0. Kilo cannot compact custom-model conversations correctly without this value.

AI Coding Tools

AI Creative Tools

Overview

Prefer VS Code?

Before You Begin

1. Install Kilo Code CLI

2. Get an EvoLink API Key

Step 1: Create the Kilo Configuration

Step 2: Validate the Configuration

Step 3: Start the Terminal Agent

Step 4: Common CLI Workflows

Troubleshooting

`kilo: command not found`

`kilo config check` reports errors

`401 unauthorized`

`404` or a duplicated request path

`model_not_found`

Chat works but file tools do not

Long sessions exceed the context window

References

​Overview

Prefer VS Code?

​Before You Begin

​1. Install Kilo Code CLI

​2. Get an EvoLink API Key

​Step 1: Create the Kilo Configuration

​Step 2: Validate the Configuration

​Step 3: Start the Terminal Agent

​Step 4: Common CLI Workflows

​Troubleshooting

​kilo: command not found

​kilo config check reports errors

​401 unauthorized

​404 or a duplicated request path

​model_not_found

​Chat works but file tools do not

​Long sessions exceed the context window

​References

Overview

Before You Begin

1. Install Kilo Code CLI

2. Get an EvoLink API Key

Step 1: Create the Kilo Configuration

Step 2: Validate the Configuration

Step 3: Start the Terminal Agent

Step 4: Common CLI Workflows

Troubleshooting

`kilo: command not found`

`kilo config check` reports errors

`401 unauthorized`

`404` or a duplicated request path

`model_not_found`

Chat works but file tools do not

Long sessions exceed the context window

References