AI/LLM

LLM

The LLM capability gives you direct access to the model client selected in bionicjs.config.ts — OpenAI, Anthropic, Google Gemini, or Ollama. The plugin hands you the official SDK client, exposed as the ai export on @bionicjs/core/server.

Providers

Each provider is a plugin package composed under the ai key:

Package	SDK	Default model
`@bionicjs/anthropic`	Anthropic SDK	`claude-sonnet-4-5`
`@bionicjs/openai`	OpenAI SDK	`gpt-4o`
`@bionicjs/google`	Google Gemini SDK	`gemini-2.5-pro`
`@bionicjs/ollama`	Ollama (local)	`llama3.2`

OpenAI

OpenAI's GPT models, wired through @bionicjs/openai. The plugin hands you the official OpenAI Python SDK client, supporting chat completions, function calling, and vision.

// bionicjs.config.ts
import { defineConfig } from "@bionicjs/core";
import { openai } from "@bionicjs/openai";

export default defineConfig({
  ai: openai({
    apiKey: process.env.OPENAI_API_KEY!,
    model: "gpt-4o",
  }),
});

OPENAI_API_KEY — server-only secret.
Default model: gpt-4o. Override it per project in bionicjs.config.ts.

Anthropic

Anthropic's Claude models, wired through @bionicjs/anthropic. Claude excels at long-context reasoning, coding, and analysis.

// bionicjs.config.ts
import { defineConfig } from "@bionicjs/core";
import { anthropic } from "@bionicjs/anthropic";

export default defineConfig({
  ai: anthropic({
    apiKey: process.env.ANTHROPIC_API_KEY!,
    model: "claude-sonnet-4-5",
  }),
});

ANTHROPIC_API_KEY — server-only secret.
Default model: claude-sonnet-4-5. Override it per project in bionicjs.config.ts.

Google Gemini

Google's Gemini models via the Google AI Python SDK, wired through @bionicjs/google. Gemini supports multimodal input (text, images, video) and has a generous free tier.

// bionicjs.config.ts
import { defineConfig } from "@bionicjs/core";
import { google } from "@bionicjs/google";

export default defineConfig({
  ai: google({
    apiKey: process.env.GOOGLE_AI_API_KEY!,
    model: "gemini-2.5-pro",
  }),
});

GOOGLE_AI_API_KEY — server-only secret.
Default model: gemini-2.5-pro. Multimodal inputs (images, video) work through the same client.

Ollama

Ollama runs models locally on your machine — no API key, no cloud dependency. It is the fastest way to develop against a real model and a good fit for privacy-sensitive workloads.

// bionicjs.config.ts
import { defineConfig } from "@bionicjs/core";
import { ollama } from "@bionicjs/ollama";

export default defineConfig({
  ai: ollama({
    baseUrl: "http://localhost:11434",
    model: "llama3.2",
  }),
});

No API key needed; ensure the Ollama daemon is running before calling the client.
Default model: llama3.2. Pull any model with ollama pull <name> and point the plugin at it.

Calling a model

The chosen provider's SDK client is exposed as ai on @bionicjs/core/server:

// server/api/chat.ts
import { ai } from "@bionicjs/core/server";

export default defineEventHandler(async (event) => {
  const { prompt } = await readBody(event);
  const response = await ai.messages.create({
    model: ai.model,
    max_tokens: 1024,
    messages: [{ role: "user", content: prompt }],
  });
  return { text: response.content[0].text };
});

On the Python side, the same client is importable directly:

# ai/llm.py
from bionicjs.ai import client

response = client.messages.create(
    model="claude-sonnet-4-5",
    max_tokens=1024,
    messages=[{"role": "user", "content": "Hello"}],
)

Good to know

The package stays thin: it records which model and API key to use and hands you the official SDK client.
Agents, RAG, tools, prompts, and MCP build on top of this client without baking combinations into the scaffold.

PreviousAgents NextMCP