Archived · v0.13.2 · Agent (Server)

Choose an LLM Provider

A2X ships three providers out of the box. Install the underlying SDK and import the corresponding A2X adapter.

Google Gemini

npm install @google/genai

import { GoogleProvider } from '@a2x/sdk/google';
 
const provider = new GoogleProvider({
  model: 'gemini-2.5-flash',
  apiKey: process.env.GOOGLE_API_KEY!,
});

Anthropic Claude

npm install @anthropic-ai/sdk

import { AnthropicProvider } from '@a2x/sdk/anthropic';
 
const provider = new AnthropicProvider({
  model: 'claude-sonnet-4-20250514',
  apiKey: process.env.ANTHROPIC_API_KEY!,
});

OpenAI GPT

npm install openai

import { OpenAIProvider } from '@a2x/sdk/openai';
 
const provider = new OpenAIProvider({
  model: 'gpt-4o',
  apiKey: process.env.OPENAI_API_KEY!,
});

Picking a model

Concern	Good default
Fast, cheap, general-purpose	`gemini-2.5-flash`, `claude-haiku-4-5`, `gpt-4o-mini`
High reasoning quality	`claude-sonnet-4-*`, `gpt-4o`, `gemini-2.5-pro`
Long-context (100k+ tokens)	any current-gen model; check the provider's docs for limits
Tool use / structured output	all three providers support tool calls; Claude and GPT tend to be the most reliable

Swap providers freely — the rest of the agent definition is identical.

Mixing providers across agents

You can use a different provider per agent in the same process. This is the usual pattern when one agent is optimized for latency and another for reasoning:

const router = new LlmAgent({
  name: 'router',
  description: 'Routes requests to a specialist agent.',
  instruction: 'Pick the best specialist.',
  provider: googleFast,           // fast classifier
  tools: [new AgentTool({ agent: analyst })],
});
 
const analyst = new LlmAgent({
  name: 'analyst',
  description: 'Deep analysis agent.',
  instruction: 'Think step by step.',
  provider: claudeSonnet,          // slower but higher-quality
});

See Multi-Agent Patterns for composition options.

Custom providers

If you need to target a provider A2X doesn't ship (e.g. a self-hosted model), implement the BaseLlmProvider interface. The shape is small — generateContent() and generateContentStream(). See the API reference for the exact contract.