Providers Guide

Configure and integrate multiple AI providers in your RelayPlane workflows. Mix and match models from OpenAI, Anthropic, Google, xAI, Perplexity, and local deployments.

OpenAI

OpenAI provides the GPT-5 series with configurable reasoning, GPT-4.1, and o-series reasoning models. The GPT-5 family features 1M token context windows and breakthrough reasoning capabilities for coding and agentic tasks.

Available Models

• •gpt-5.2 - New! Latest flagship with enhanced reasoning and 1M context
• •gpt-5.2-pro - Smarter and more precise responses for complex tasks
• •gpt-5.1 - Best model for coding and agentic tasks with configurable reasoning
• •gpt-5-mini - Faster, cost-efficient version of GPT-5 for well-defined tasks
• •gpt-5-nano - Fastest, most cost-efficient version of GPT-5
• •gpt-5 - Previous intelligent reasoning model
• •gpt-4.1 - Smartest non-reasoning model with 1M context
• •o3 - Advanced reasoning model
• •o4-mini - Fast reasoning model

Configuration

1import { relay } from '@relayplane/sdk'
2
3// Configure your provider (do this once at app startup)
4relay.configure({
5  providers: {
6    openai: { apiKey: process.env.OPENAI_API_KEY! }
7  }
8})
9
10const workflow = relay
11  .workflow('openai-example')
12  .step('analyze')
13  .with('openai:gpt-5.2')
14  .prompt('Analyze the provided text and extract key insights: {{input.text}}')
15
16const result = await workflow.run({ text: 'Your content here...' })

Best Use Cases

• •GPT-5.2: Latest flagship for complex reasoning, coding, and agentic tasks
• •GPT-5.2 Pro: Most precise responses, research, complex analysis
• •GPT-5 Nano: High-volume classification, simple transformations
• •GPT-4.1: General purpose, coding without reasoning overhead

Anthropic

Anthropic's Claude 4 series features hybrid thinking modes and industry-leading coding performance. Claude Opus 4.5 is the most intelligent, while Sonnet 4.5 offers the best balance of capability and speed.

Available Models

• •claude-opus-4-5-20251101 - Most intelligent, effort parameter for complex tasks
• •claude-sonnet-4-5-20250929 - Best coding model, strongest for agents
• •claude-haiku-4-5-20251001 - Fast and affordable for high-volume tasks
• •claude-opus-4-1-20250805 - Superior precision for agentic tasks
• •claude-sonnet-4-20250514 - Upgraded reasoning with hybrid thinking
• •claude-3-7-sonnet-20250219 - Extended thinking capabilities
• •claude-3-5-haiku-20241022 - Fast and affordable (legacy)

Configuration

1import { relay } from '@relayplane/sdk'
2
3// Configure your provider (do this once at app startup)
4relay.configure({
5  providers: {
6    anthropic: { apiKey: process.env.ANTHROPIC_API_KEY! }
7  }
8})
9
10const analysisPrompt = `You are an expert analyst. Thoroughly analyze
11the following document and provide detailed insights with
12supporting evidence from the text.
13
14Document: {{input.document}}`
15
16const workflow = relay
17  .workflow('anthropic-example')
18  .step('deep-analysis')
19  .with('anthropic:claude-sonnet-4-5-20250929')
20  .prompt(analysisPrompt)
21
22// longDocument can be up to 200k tokens
23const result = await workflow.run({ document: longDocument })

Best Use Cases

• •Opus 4.5: Most complex reasoning, research, strategic analysis
• •Sonnet 4.5: Coding, agentic workflows, computer use
• •Claude 3.5 Haiku: Fast preprocessing, high-volume tasks

Google (Gemini)

Google's Gemini 3 and 2.5 series offer massive 1M+ token context windows and multimodal capabilities including text, image, video, audio, and PDF processing.

Available Models

• •gemini-3-pro - Most powerful Gemini with text, image, video, audio, PDF
• •gemini-2.5-pro - Advanced multimodal with 1M context
• •gemini-2.5-flash - Fast multimodal for general tasks
• •gemini-2.5-flash-lite - Ultra-efficient for high-frequency tasks
• •gemini-2.0-flash - Cost-effective multimodal

Configuration

1import { relay } from '@relayplane/sdk'
2
3// Configure your provider (do this once at app startup)
4relay.configure({
5  providers: {
6    google: { apiKey: process.env.GOOGLE_API_KEY! }
7  }
8})
9
10const workflow = relay
11  .workflow('google-example')
12  .step('process')
13  .with('google:gemini-2.5-flash')
14  .prompt('Process the input and provide a structured response: {{input.content}}')
15
16const result = await workflow.run({ content: 'Your content here...' })

Best Use Cases

• •Video/Audio Processing: Gemini 3 Pro supports native video and audio
• •Long Documents: 1M+ context for entire codebases or books
• •Cost Optimization: Gemini 2.5 Flash-Lite at $0.02/1M input tokens

xAI (Grok)

xAI's Grok 4 models feature 256K context windows and advanced reasoning with planning capabilities. Grok 4 excels at real-time knowledge tasks.

Available Models

• •grok-4 - Latest flagship with reasoning and planning (256K context)
• •grok-4-fast - Faster variant of Grok 4
• •grok-3 - Previous generation flagship
• •grok-3-mini - Smaller, faster Grok for quick tasks

Configuration

1import { relay } from '@relayplane/sdk'
2
3// Configure your provider (do this once at app startup)
4relay.configure({
5  providers: {
6    xai: { apiKey: process.env.XAI_API_KEY! }
7  }
8})
9
10const workflow = relay
11  .workflow('xai-example')
12  .step('research')
13  .with('xai:grok-4')
14  .prompt('Research the topic and provide current information: {{input.topic}}')
15
16const result = await workflow.run({ topic: 'Latest developments in AI' })

Best Use Cases

• •Real-time Information: Current events, trending topics, live data
• •Reasoning Tasks: Complex planning and analysis
• •Creative Content: Less restrictive content generation

Perplexity

Perplexity's Sonar models combine language understanding with real-time web search capabilities, making them ideal for research and information retrieval tasks.

Available Models

• •sonar-pro - Most capable Sonar model with advanced search
• •sonar - Balanced performance and cost
• •sonar-reasoning-pro - Enhanced reasoning with web search
• •sonar-reasoning - Reasoning capabilities with search
• •sonar-deep-research - Comprehensive research mode

Configuration

1import { relay } from '@relayplane/sdk'
2
3// Configure your provider (do this once at app startup)
4relay.configure({
5  providers: {
6    perplexity: { apiKey: process.env.PERPLEXITY_API_KEY! }
7  }
8})
9
10const workflow = relay
11  .workflow('perplexity-example')
12  .step('research')
13  .with('perplexity:sonar-pro')
14  .prompt('Research the following topic with current information: {{input.topic}}')
15
16const result = await workflow.run({ topic: 'Latest AI developments 2025' })

Best Use Cases

• •Real-time Research: Topics requiring up-to-date web information
• •Fact Checking: Verify claims against current sources
• •Competitive Analysis: Market research with citations

Local (Ollama)

Run models locally using Ollama for privacy-sensitive data, offline operation, or cost optimization on high-volume tasks.

Setup Instructions

First, install Ollama and pull your desired model:

1# Install Ollama (macOS/Linux)
2curl -fsSL https://ollama.com/install.sh | sh
3
4# Pull models
5ollama pull llama3.3
6ollama pull qwen2.5
7ollama pull deepseek-r1
8ollama pull mistral
9
10# Start Ollama server (runs on http://localhost:11434)
11ollama serve

Available Models

• •llama3.3 - Meta's latest 70B open model with vision
• •llama3.2 - Multimodal open-source model
• •qwen2.5 - Alibaba's powerful open model
• •deepseek-r1 - Advanced reasoning model
• •mistral - Fast and efficient for many tasks

Configuration

1import { relay } from '@relayplane/sdk'
2
3// No API key needed for local models, but you can set custom endpoint
4relay.configure({
5  providers: {
6    local: { baseUrl: 'http://localhost:11434' } // Optional: customize Ollama endpoint
7  }
8})
9
10const workflow = relay
11  .workflow('local-example')
12  .step('process')
13  .with('local:llama3.3')
14  .prompt('Process the sensitive data locally: {{input.data}}')
15
16const result = await workflow.run({ data: sensitiveData })

Multi-Provider Configuration

RelayPlane makes it easy to use multiple providers in a single workflow, allowing you to leverage the strengths of each model.

Multi-Provider Workflow Example

1import { relay } from '@relayplane/sdk'
2
3// Configure all providers once at app startup
4relay.configure({
5  providers: {
6    openai: { apiKey: process.env.OPENAI_API_KEY! },
7    anthropic: { apiKey: process.env.ANTHROPIC_API_KEY! }
8  }
9})
10
11const multiProviderWorkflow = relay
12  .workflow('multi-provider-analysis')
13
14  // Step 1: Fast initial processing with GPT-5-nano
15  .step('extract')
16  .with('openai:gpt-5-nano')
17  .prompt('Extract key data points from the document: {{input.document}}')
18
19  // Step 2: Deep analysis with Claude Sonnet 4.5
20  .step('analyze')
21  .with('anthropic:claude-sonnet-4-5-20250929')
22  .depends('extract')
23  .prompt('Perform deep analysis on the extracted data: {{extract.output}}')
24
25  // Step 3: Generate report with GPT-5.2
26  .step('report')
27  .with('openai:gpt-5.2')
28  .depends('analyze')
29  .prompt('Generate a professional report based on: {{analyze.output}}')
30
31  // Step 4: Privacy-sensitive summary with local model
32  .step('internal-summary')
33  .with('local:llama3.3')
34  .depends('analyze')
35  .prompt('Create an internal summary with confidential notes based on: {{analyze.output}}')
36
37const result = await multiProviderWorkflow.run({ document: documentContent })

Cost Comparison

Understanding pricing helps you optimize workflows for cost-effectiveness. Prices are per 1M tokens (as of December 2025).

Token Pricing by Provider

Model Capability Matrix

Next Steps

• •Quickstart Guide - Build your first workflow
• •API Reference - Complete SDK documentation
• •Example Workflows - Production-ready templates
• •Core Concepts - DAG validation, error handling, retries