Kilo Code - Capriole AI API

What is Kilo Code

Kilo Code is an AI coding assistant for IDEs and terminals. It can use Capriole AI through Kilo’s OpenAI-compatible provider, which sends requests to Capriole AI’s Chat Completions-compatible API.

Prerequisites

Before you begin, make sure you have:

A Capriole AI account (Sign up here)
A Capriole AI API key (Get your key)
Kilo Code installed in VS Code or as the Kilo CLI

Installation

VS Code extension

Install Kilo Code from the VS Code Extensions view. Search for Kilo Code, then install the pre-release version.

Kilo CLI

npm install -g @kilocode/cli

Configuration

Create or update your Kilo config file:

Global config: ~/.config/kilo/kilo.jsonc
Project config: kilo.jsonc or .kilo/kilo.jsonc in your project root

Use this OpenAI-compatible provider setup:

{
  "$schema": "https://app.kilo.ai/config.json",
  "model": "openai-compatible/openai-latest",
  "enabled_providers": ["openai-compatible"],
  "provider": {
    "openai-compatible": {
      "options": {
        "apiKey": "{env:CAPRIOLE_AI_API_KEY}",
        "baseURL": "https://api.caprioletech.com/v1"
      },
      "models": {
        "openai-latest": {
          "name": "Capriole OpenAI Latest",
          "tool_call": true,
          "reasoning": true
        },
        "claude-latest": {
          "name": "Capriole Claude Latest",
          "tool_call": true,
          "reasoning": true
        },
        "anthropic/claude-fable-5": {
          "name": "Capriole Fable 5",
          "tool_call": true,
          "reasoning": true
        },
        "google-latest": {
          "name": "Capriole Google Latest",
          "tool_call": true,
          "reasoning": true
        },
        "openai/gpt-5.6-terra": {
          "name": "Capriole GPT 5.6 Thinking",
          "tool_call": true,
          "reasoning": true
        },
        "openai/gpt-5.6-luna": {
          "name": "Capriole GPT 5.6",
          "tool_call": true,
          "reasoning": true
        },
        "openai/gpt-5.5": {
          "name": "Capriole GPT 5.5",
          "tool_call": true,
          "reasoning": true
        },
        "openai/gpt-5.4-mini": {
          "name": "Capriole GPT 5.4 Mini",
          "tool_call": true,
          "reasoning": true
        },
        "xai/grok-4.5": {
          "name": "Capriole Grok 4.5",
          "tool_call": true,
          "reasoning": true
        },
        "zai/glm-5.2": {
          "name": "Capriole GLM 5.2",
          "tool_call": true,
          "reasoning": true
        },
        "anthropic/claude-opus-4-8": {
          "name": "Capriole Opus 4.8",
          "tool_call": true
        },
        "anthropic/claude-opus-4-7": {
          "name": "Capriole Opus 4.7",
          "tool_call": true
        },
        "anthropic/claude-opus-4-6": {
          "name": "Capriole Opus 4.6",
          "tool_call": true
        },
        "anthropic/claude-sonnet-4-6": {
          "name": "Capriole Sonnet 4.6",
          "tool_call": true
        },
        "google/gemini-3.1-pro-preview": {
          "name": "Capriole Gemini 3.1 Pro Preview",
          "tool_call": true,
          "reasoning": true
        },
        "google/gemini-3.5-flash": {
          "name": "Capriole Gemini 3.5 Flash",
          "tool_call": false,
          "reasoning": true
        }
      }
    }
  }
}

Run Kilo Code

Set your API key, then start Kilo from your project directory:

export CAPRIOLE_AI_API_KEY="YOUR_CAPRIOLE_AI_API_KEY"
kilo

Start with a specific model:

kilo -m openai-compatible/openai-latest
kilo -m openai-compatible/claude-latest
kilo -m openai-compatible/google-latest
kilo -m openai-compatible/openai/gpt-5.5
kilo -m openai-compatible/openai/gpt-5.4-mini
kilo -m openai-compatible/anthropic/claude-fable-5
kilo -m openai-compatible/anthropic/claude-opus-4-8
kilo -m openai-compatible/anthropic/claude-opus-4-7
kilo -m openai-compatible/google/gemini-3.1-pro-preview
kilo -m openai-compatible/google/gemini-3.5-flash

Run a one-off request with Gemini 3.5 Flash:

kilo run --agent ask -m openai-compatible/google/gemini-3.5-flash "Summarize this project."

For a one-off CLI request, pass a prompt to kilo run:

export CAPRIOLE_AI_API_KEY="YOUR_CAPRIOLE_AI_API_KEY"
kilo run -m openai-compatible/openai-latest "Explain this repository structure."

Supported models

Start with openai-latest for new Kilo Code setups. Use claude-latest or google-latest when you want Capriole AI to route to our recommended flagship Claude or Gemini model, and use a concrete model ID when you need an exact model version. The Kilo docs use the same family order as the Capriole AI model picker: OpenAI, Anthropic, then Google. Public API integrations exclude web-chat-only presets such as Fable 5 Thinking and Opus 4.8 Thinking.

openai-latest
openai/gpt-5.6-terra
openai/gpt-5.6-luna
openai/gpt-5.5
openai/gpt-5.4-mini
claude-latest
anthropic/claude-fable-5
anthropic/claude-opus-4-8
anthropic/claude-opus-4-7
anthropic/claude-opus-4-6
anthropic/claude-sonnet-4-6
google-latest
google/gemini-3.1-pro-preview
google/gemini-3.5-flash
xai/grok-4.5
zai/glm-5.2

Keep "tool_call": false for Gemini 3.5 Flash in Kilo Code. Capriole AI supports simple Chat Completions tool calls for this model, and Kilo Code can run local tool workflows with this setting, but enabling native Gemini function declarations can still expose Gemini’s narrower OpenAPI schema subset. Other public Chat Completions-compatible model IDs returned by GET /v1/models can also be configured when you want to test them. Validate tool use before using a model for code-editing workflows.

Troubleshooting

Why does Kilo Code return an authentication error?

Confirm that CAPRIOLE_AI_API_KEY is exported in the same shell before running kilo.

Which config file should I edit?

Use ~/.config/kilo/kilo.jsonc for the current Kilo CLI and VS Code extension. Run kilo config check after editing it.

Why do requests fail with endpoint errors?

Confirm that baseURL is https://api.caprioletech.com/v1. Kilo sends Chat Completions requests to /v1/chat/completions.

Why does Kilo Code say the model is missing?

Confirm that the selected model uses the openai-compatible/<model-id> format and that provider.openai-compatible.models contains the matching model entry. For example, openai-compatible/openai-latest maps to the openai-latest config key.

Why does tool use fail?

For GPT, Claude, Gemini 3.1 Pro, and latest-alias entries, confirm that the model entry includes "tool_call": true. Kilo Code executes tools locally and sends follow-up Chat Completions requests to Capriole AI. For Gemini 3.5 Flash, keep "tool_call": false. Gemini supports function calling, but its function declarations use a supported subset of OpenAPI schema. Some native tool schemas are broader than that subset.

Notes

Capriole AI exposes Kilo Code through POST /v1/chat/completions.
Kilo Code executes coding tools locally. Capriole AI only handles the model request.
Keep the provider config minimal first. Add Kilo model limits or other model options only when you intentionally want client-side context management.
Official references: Kilo OpenAI Compatible provider, Kilo settings, Kilo custom models, Kilo CLI, and Gemini function calling.

​What is Kilo Code

​Prerequisites

​Installation

​VS Code extension

​Kilo CLI

​Configuration

​Run Kilo Code

​Supported models

​Troubleshooting

​Why does Kilo Code return an authentication error?

​Which config file should I edit?

​Why do requests fail with endpoint errors?

​Why does Kilo Code say the model is missing?

​Why does tool use fail?

​Notes