Connect Kilo Code

Current Kilo Code supports custom providers in its editor and CLI. Use the route for your installed client; older extension versions can have different settings.

Prepare a service key, an exact model ID from our models endpoint, and verified context/output limits. Choose a model with streaming and tool calling for agent work.

Configure the editor

Open Kilo Code’s Settings → Providers, scroll down and click Custom provider.

Use the API base without /chat/completions. Enter the key in Kilo’s key field. This guide selects the Chat Completions format; choose a model that supports that format.

Kilo can fetch /v1/models after the URL and key are entered. Use Fetch models to retry, or add an exact ID manually. Click Submit, then choose my-api/YOUR_MODEL_ID in the model picker. The first part is Kilo’s local provider ID, while the service receives the underlying model ID. See Kilo’s compatible-provider guide.

Configure Kilo CLI with a separate file

For an agent-assisted setup, create an unused file such as ~/.config/kilo/service.jsonc. If XDG_CONFIG_HOME is set, use its kilo directory instead. Preserve existing configuration and tool permissions. KILO_CONFIG adds to the normal configuration; it does not isolate the launch from project or managed settings.

Replace YOUR_MODEL_ID with the exact service model ID. The two [TBD] values are required model facts: replace the quoted strings with verified integer limits before using the file.

{
  "$schema": "https://app.kilo.ai/config.json",
  "provider": {
    "my-api": {
      "npm": "@ai-sdk/openai-compatible",
      "name": "My API",
      "options": {
        "baseURL": "https://aelder.dev/v1",
        "apiKey": "{env:SERVICE_API_KEY}"
      },
      "models": {
        "YOUR_MODEL_ID": {
          "name": "My API model",
          "tool_call": true,
          "modalities": {
            "input": ["text"],
            "output": ["text"]
          },
          "limit": {
            "context": "[TBD]",
            "output": "[TBD]"
          }
        }
      }
    }
  }
}

Kilo documents these fields in Custom Models and its configuration schema. Unknown custom-model limits can disable automatic context management; obtain the real values before starting a long task. Keep tool_call: true only for a model with verified tool support.

In Bash or Zsh, enter your key at a hidden prompt and launch the CLI:

printf 'Service API key: '
IFS= read -r -s SERVICE_API_KEY
printf '\n'
export SERVICE_API_KEY

KILO_CONFIG="${XDG_CONFIG_HOME:-$HOME/.config}/kilo/service.jsonc" \
kilo --model "my-api/YOUR_MODEL_ID"

unset SERVICE_API_KEY

Replace the model ID in the command as well. A file selected through KILO_CONFIG is trusted configuration, so Kilo can resolve its {env:SERVICE_API_KEY} reference. A repository-level config may not resolve secret references. The CLI configuration guide explains this behavior and the required restart after file changes.

Verify the active model

Start a new conversation and ask for CONNECTION_OK without tools. Check the selected model, the reply and a matching request in your service dashboard. Then try a small file-reading task with normal approvals to verify tool calls. Each prompt uses API balance.

Listing models or saving a provider only confirms configuration. If the wrong connection is active, check the chosen model and any project, organization or managed configuration that overrides the file. For errors, check /v1, the key and exact ID before using our API troubleshooting. A custom model connection does not configure separate cloud features or embeddings.