Kimi Code¶

Autobot supports Kimi Code (by Moonshot AI) as an LLM provider. Kimi Code is highly optimized for coding tasks and agentic workflows.

Setup¶

1. Get an API key¶

Create an API key on the Kimi Code membership page.

2. Configure credentials¶

Add your API key to the .env file:

KIMI_API_KEY=sk-kimi-...

Or use the interactive setup:

autobot setup
# Select "Kimi Code" as provider

3. Configure the provider¶

In config.yml:

agents:
  defaults:
    model: "kimi/kimi-for-coding"

providers:
  kimi:
    api_key: "${KIMI_API_KEY}"

4. Verify¶

autobot doctor
# Should show: ✓ LLM provider configured (kimi)

Model naming¶

Models use the kimi/ prefix followed by the Kimi model ID:

model: "kimi/kimi-for-coding"   # Optimized for coding tasks (recommended)

The kimi/ prefix tells autobot to route to the Kimi Code API. It is stripped before sending to the API.

Configuration reference¶

Field	Required	Default	Description
`api_key`	Yes	—	Kimi API key (`sk-kimi-...`)
`api_base`	No	`https://api.kimi.com/coding/v1/chat/completions`	Custom API endpoint
`extra_headers`	No	—	Additional HTTP headers for every request

How it works¶

Kimi Code uses an OpenAI-compatible Chat Completions API. Autobot detects Kimi models by the kimi/ prefix or keywords and routes to the coding-optimized endpoint automatically.

Context Window: 262,144 tokens
Max Output: 32,768 tokens
Optimization: Highly tuned for tool use and long-context coding tasks.

Troubleshooting¶

Enable debug logging to see request/response details:

LOG_LEVEL=DEBUG autobot agent -m "Hello"

Look for:

POST https://api.kimi.com/coding/v1/chat/completions model=... — confirms provider is active
Response 200 (N bytes) — confirms API response
HTTP 4xx/5xx: ... — API errors with details

Common issues¶

"API error: Rate limit reached" — Too many requests or tokens. Check your Kimi account status.

"API error: The model does not exist" — Model ID is wrong. Ensure you are using kimi-for-coding.