Kilo Code with mittwald
Kilo Code is a Visual Studio Code extension that provides AI-powered coding assistance with support for OpenAI-compatible providers.
Installation
Install the Kilo Code extension from the VS Code Marketplace.
Configuration
Configure Kilo Code with mittwald's AI models:
- Open Kilo Code settings in VS Code (gear icon) and go to the Providers tab
- At the bottom of the list, click Custom provider
- Fill in the dialog:
- Provider ID and Display name:
mittwald - Provider API: OpenAI Compatible
- Base URL:
https://llm.aihosting.mittwald.de/v1 - API key: Your mittwald AI Hosting API key (how to get one)
- Models:
DeepSeek-V4.1-Flash-1017
- Provider ID and Display name:
- Click Submit
DeepSeek-V4.1-Flash-1017 has a context window of 1,048,576 tokens, reads images, and reasons
before it answers. Token limits and image input are set in the kilo.jsonc configuration
file, see the Kilo Code guide to custom models. Set the context limit to 1048576,
the output limit to at least 32768, and add image to the input modalities.
The model is in the Testphase. For production use, choose Qwen3.8-27B-NVFP4
(Stable) instead, with a context limit of 256000.
Environment Variables
Alternatively, configure via environment variables:
user@local $ export OPENAI_API_KEY=sk-your-api-key-here
user@local $ export OPENAI_BASE_URL=https://llm.aihosting.mittwald.de/v1
Then select the model in Kilo Code's model picker for agent tasks.
MCP Server Integration
Kilo Code supports MCP servers. Add them via settings (Tools → Add MCP Server) or in the configuration file:
{
"mcpServers": {
"context7": {
"command": "npx",
"args": ["-y", "@upstash/context7-mcp"],
"env": { "CONTEXT7_API_KEY": "your-key" }
},
"brave-search": {
"command": "npx",
"args": ["-y", "@brave/brave-search-mcp-server"],
"env": { "BRAVE_API_KEY": "your-key" }
},
"directus": {
"command": "npx",
"args": ["-y", "@directus/content-mcp", "--stdio"],
"env": {
"DIRECTUS_URL": "https://your-directus-instance.com",
"DIRECTUS_TOKEN": "your-directus-token"
}
},
"opensearch": {
"command": "uvx",
"args": ["opensearch-mcp-server-py"],
"env": {
"OPENSEARCH_URL": "https://localhost:9200",
"OPENSEARCH_USERNAME": "your-opensearch-username",
"OPENSEARCH_PASSWORD": "your-opensearch-password"
}
},
"typo3": {
"command": "mw",
"args": [
"app",
"exec",
"--quiet",
"--installation-id=a-XXXXX",
"vendor/bin/typo3 mcp:server"
]
},
"mittwald": {
"transport": {
"type": "sse",
"url": "https://mcp.mittwald.de/mcp"
}
}
}
}
Web Fetching
Kilo Code can use web-fetch/browsing tools to retrieve and summarize external pages:
Best practices:
- Prefer public URLs (no auth)
- Provide exact target links in your prompt
- Keep network scope small
- Ask the agent to include citation URLs next to each claim
- Use deterministic sampling for factual extraction:
temperature <= 0.2
Example prompts:
- "Fetch https://example.com/spec and summarize key endpoints in 10 bullet points. Include canonical links."
- "Open https://huggingface.co/model-page and extract context length, modalities, and license. Return as JSON."
Next Steps
- Try other integrations - Compare different tools
- Learn about MCP - Infrastructure management
- Best practices - Optimize your AI usage