@kb-labs/adapters-vibeproxy

Part of KB Labs ecosystem. Works exclusively within KB Labs platform.

VibeProxy local adapter supporting multiple LLM providers (Claude, GPT, etc.) through a unified interface.

Overview

| Property | Value | |----------|-------| | Implements | ILLM | | Type | core | | Requires | None | | Category | AI |

Features

• Multi-Provider - Claude, GPT, Gemini, and more via single interface
• Local Proxy - Route through local VibeProxy server
• Function Calling - Native tool support for all providers
• Model Switching - Change provider by just changing model name

Installation

pnpm add @kb-labs/adapters-vibeproxy

Configuration

Add to your kb.config.json:

{
  "platform": {
    "adapters": {
      "llm": "@kb-labs/adapters-vibeproxy"
    },
    "adapterOptions": {
      "llm": {
        "baseURL": "http://localhost:8317",
        "apiKey": "any-string",
        "model": "claude-sonnet-4-20250514",
        "timeout": 120000
      }
    }
  }
}

Options

| Option | Type | Default | Description | |--------|------|---------|-------------| | baseURL | string | "http://localhost:8317" | VibeProxy server URL | | apiKey | string | "any-string" | API key (any string works for local) | | model | string | "claude-sonnet-4-20250514" | Model to use | | timeout | number | 120000 | Request timeout in ms |

Usage

Via Platform (Recommended)

import { usePlatform } from '@kb-labs/sdk';

const platform = usePlatform();

// Chat with Claude
const response = await platform.llm.chat([
  { role: 'user', content: 'Hello!' }
]);

// Switch to GPT by changing model
const gptResponse = await platform.llm.chat(
  [{ role: 'user', content: 'Hello!' }],
  { model: 'gpt-4-turbo' }
);

// Function calling
const result = await platform.llm.chatWithTools(
  [{ role: 'user', content: 'What time is it?' }],
  [{ name: 'getTime', parameters: { ... } }]
);

Standalone (Testing/Development)

import { createAdapter } from '@kb-labs/adapters-vibeproxy';

const llm = createAdapter({
  baseURL: 'http://localhost:8317',
  model: 'claude-sonnet-4-20250514'
});

const response = await llm.chat([
  { role: 'user', content: 'Hello!' }
]);

Adapter Manifest

{
  id: 'vibeproxy-llm',
  name: 'VibeProxy LLM',
  version: '0.1.0',
  implements: 'ILLM',
  capabilities: {
    streaming: false, // TODO: implement SSE streaming
    custom: {
      functionCalling: true,
      multiProvider: true,
    },
  },
}

Supported Models

| Provider | Model Examples | |----------|---------------| | Anthropic | claude-sonnet-4-20250514, claude-3-opus-* | | OpenAI | gpt-4-turbo, gpt-3.5-turbo | | Google | gemini-pro, gemini-ultra |

FAQ

See VibeProxy documentation for setup instructions. Default port is 8317.

• Single interface for multiple providers
• Local caching and rate limiting
• Request logging and analytics
• Cost tracking across providers

Not yet. Streaming (SSE) is planned for a future release.

Related Adapters

| Adapter | Use Case | |---------|----------| | @kb-labs/adapters-openai | Direct OpenAI API access |

License

KB Public License v1.1 - KB Labs Team