Kilo Code

Checked against Kilo Code's docs on 2026-10-11

Kilo Code connects as a custom provider: one block in ~/.config/kilo/kilo.jsonc, which the VS Code extension and the kilo command share.

Base URL: https://api.creeksidecompute.ai/v1
API: OpenAI Chat Completions
Model id: qwen/qwen3.8-27b, or qwen/qwen3.6-35b-a3b for speed, or meta/muse-glimmer-30b, which always reasons first
Context window: 131072 tokens, up to 65536 of them output
Key: Make one at creeksidecompute.ai/account

Set it up

Make a key. Sign in at creeksidecompute.ai/account with your email and make a key. Each request is paid from prepaid credit: top up by card from $10, or redeem a promotional code.
Put the key in an environment variable. Add this line to your shell profile (for example ~/.zshrc), then restart VS Code:
shell
```
export CREEKSIDE_API_KEY=<your key>
```

Add the provider. Put this in ~/.config/kilo/kilo.jsonc. Keep it in this global file: Kilo reads {env:…} only there, not in a project's kilo.json.

~/.config/kilo/kilo.jsonc

{
  "$schema": "https://app.kilo.ai/config.json",
  "model": "creekside/qwen/qwen3.8-27b",
  "small_model": "creekside/qwen/qwen3.8-27b",
  "provider": {
    "creekside": {
      "npm": "@ai-sdk/openai-compatible",
      "name": "Creekside",
      "options": {
        "baseURL": "https://api.creeksidecompute.ai/v1",
        "apiKey": "{env:CREEKSIDE_API_KEY}"
      },
      "models": {
        "qwen/qwen3.8-27b": {
          "name": "Qwen3.8 27B",
          "tool_call": true,
          "reasoning": true,
          "limit": { "context": 131072, "output": 32768 },
          "variants": {
            "none": { "reasoningEffort": "none" },
            "low": { "reasoningEffort": "low" },
            "medium": { "reasoningEffort": "medium" },
            "high": { "reasoningEffort": "high" }
          }
        },
        "qwen/qwen3.6-35b-a3b": {
          "name": "Qwen3.6 35B A3B",
          "tool_call": true,
          "reasoning": true,
          "limit": { "context": 131072, "output": 32768 },
          "variants": {
            "none": { "reasoningEffort": "none" },
            "low": { "reasoningEffort": "low" },
            "medium": { "reasoningEffort": "medium" },
            "high": { "reasoningEffort": "high" }
          }
        },
        "meta/muse-glimmer-30b": {
          "name": "Muse Glimmer 30B",
          "tool_call": true,
          "reasoning": true,
          "attachment": true,
          "modalities": { "input": ["text", "image"], "output": ["text"] },
          "limit": { "context": 131072, "output": 32768 },
          "variants": {
            "low": { "reasoningEffort": "low" },
            "medium": { "reasoningEffort": "medium" },
            "high": { "reasoningEffort": "high" }
          }
        }
      }
    }
  }
}

small_model keeps Kilo's small requests, such as session titles, on Creekside. Without limit, Kilo never compacts, and the conversation grows until a request is refused. In VS Code, Settings, Providers, Custom provider adds the same provider from a form and stores the key for you. It sets no limits or variants, so add the models block as well. attachment and modalities let Kilo send images to Muse Glimmer 30B, which reads them.

Pick the model. In VS Code, choose Qwen3.8 27B under Creekside in the model picker, or type /models. In the terminal: kilo -m creekside/qwen/qwen3.8-27b. For speed, choose Qwen3.6 35B A3B instead, or kilo -m creekside/qwen/qwen3.6-35b-a3b. For Muse Glimmer 30B, choose it under Creekside, or kilo -m creekside/meta/muse-glimmer-30b.

Thinking

Thinking is off until you pick a variant: with none selected, Kilo sends no thinking setting. The four variants send reasoning_effort: none keeps thinking off, and low, medium and high turn it on. Type /variant in VS Code or /variants in the CLI to choose one, or press Shift+Tab in the prompt to cycle through them. Thinking tokens are billed as output.

Muse Glimmer 30B always reasons before it answers, so its variants set how long, not whether: with no variant selected, or at low, it reasons briefly, and medium and high make it reason longer.

What does not work

Images and browser use, on the Qwen models. They are listed for text only, so they cannot read the screenshots that browser use returns. Muse Glimmer 30B's entry takes images.
Cost display. Without a cost block, Kilo shows no spend. Your bill uses the prices in GET /v1/models.

Troubleshooting

401 invalid API key: The key is wrong, or Kilo Code did not get it. Set CREEKSIDE_API_KEY in your shell profile and restart VS Code, or enter the key again in the Custom provider form. To test the key alone, curl -s https://api.creeksidecompute.ai/v1/models -H "Authorization: Bearer $CREEKSIDE_API_KEY" returns the model list when the key is good.
404 model … is not served: The model id is wrong. Use qwen/qwen3.8-27b exactly, in lower case, with the qwen/ prefix. GET /v1/models lists every id we serve. In kilo.jsonc the key under models is the id; creekside/ is only Kilo's provider prefix.
400 maximum context length is 131072 tokens: The conversation is longer than the model's window. Keep limit.context at 131072 so Kilo compacts in time, then start a new task.
402 or 429: A 402 means your prepaid credit does not cover the request: add credit at creeksidecompute.ai/account. A 429 means the key's rate, or the number of requests your account may run at once, is used up; agents that work in parallel reach that sooner. Wait and retry. The API reference lists the limits.

Other tools

Sources, read on 2026-10-11: OpenAI-compatible providers, custom models, model selection.