Rueter AI

Rueter AI is a TypeScript-first ESM library that gives Anthropic, OpenAI, Google Gemini, xAI Grok, and DeepSeek a consistent prompt interface.

The two core primitives are:

• RueterModel for a single provider/model instance
• Rueter for parallel orchestration across multiple RueterModel instances

On top of that core, the package also exports:

• 53 preconfigured specialized model factories
• 5 higher-level workflows
• GhostWriter, a convenience helper for writing-style replication

What It Does

• Normalizes provider request shapes behind a single .prompt() API
• Returns plain text or structured result objects with usage cost data
• Lets you fan the same prompt out to multiple models in parallel
• Ships with many task-specific model presets for code, planning, writing, data extraction, and more
• Uses native fetch under the hood instead of provider SDKs
• Exports public TypeScript types for the main result/config shapes

Runtime Requirements

• Node.js 18+ or any runtime with a global fetch
• ESM project setup
• API keys for the providers you plan to call

Installation

npm install rueter-ai

Global CLI install:

sudo npm i -g rueter-ai

Use whatever environment-loading strategy your app already prefers. The examples below assume API keys are already available on process.env.

CLI

The package now ships a global rueter terminal command.

Common commands:

rueter doctor
rueter models catalog
rueter models create
rueter models run my-model --prompt "Explain optimistic locking."
rueter orchestrators create
rueter presets list
rueter workflows list
rueter ghostwriter run --history-file ./samples.txt --prompt "Write a project update."
rueter history list

The CLI automatically loads environment variables from a local .env file via dotenv when you run it in a project directory. Prompt-producing commands record local run history under .rueter/history/ by default; pass --no-history for private one-off runs.

Quick Start

Single model

import { RueterModel } from "rueter-ai"

const model = new RueterModel("openai", process.env.OPENAI_API_KEY, 2, {
  systemPrompt: "You are concise and technical.",
  temperature: 0.2,
  maxTokens: 512,
})

const text = await model.prompt("What is the capital of Japan?")
console.log(text)

const detailed = await model.prompt("What is the capital of Japan?", true)
console.log(detailed)

model.prompt(prompt) returns a string.

model.prompt(prompt, true) returns:

{
  res: string | null
  cost: {
    model: string
    input: string
    output: string
    total: string
  } | null
  error?: string
}

Multi-model orchestration

import { Rueter, RueterModel } from "rueter-ai"

const models = [
  new RueterModel("grok", process.env.GROK_API!, 1, {
    temperature: 0,
    maxTokens: 512,
  }),
  new RueterModel("anthropic", process.env.ANTHROPIC_API_KEY!, 2, {
    temperature: 0,
    maxTokens: 512,
  }),
]

const rueter = new Rueter(models, {
  systemPrompt: "Answer briefly and directly.",
  temperature: 0,
  maxTokens: 512,
})

const results = await rueter.prompt("What is the capital of Japan?")
console.log(results)

rueter.prompt(prompt) returns a Record<string, ModelResult> keyed by each model's runtime ID, such as openai_gpt-5.4 or grok_grok-4-1-fast-reasoning.

Public API

RueterModel

Constructor:

new RueterModel(provider, apiKey, modelIndex?, config?)

Parameters:

• provider: "anthropic" | "openai" | "gemini" | "grok" | "deepseek"
• apiKey: provider API key
• modelIndex: numeric index into the built-in model catalog for that provider, default 0
• config: optional model settings

Supported config fields:

• systemPrompt?: string
• temperature?: number
• maxTokens?: number
• stopSequences?: string[]

Instance methods:

• prompt(prompt: string): Promise<string>
• prompt(prompt: string, returnJSON: true): Promise<ModelResult>
• setSystemPrompt(sysPrompt: string): void
• setTemperature(temp: number): void
• setMaxTokens(maxTok: number): void
• getID(): string

Rueter

Constructor:

new Rueter(models, config)

Parameters:

• models: RueterModel[]
• config: shared config applied across the supplied models

Important: the second argument is required by the current implementation. If you do not want shared overrides, pass {}.

Orchestrator methods:

• prompt(prompt: string): Promise<RueterResults>
• addModel(provider, apiKey, modelIndex?): void
• setSystemPrompt(sysPrompt: string): void
• setTemperature(temp: number): void
• setMaxTokens(maxTok: number): void

Current orchestrator-wide shared settings are limited to:

• systemPrompt
• temperature
• maxTokens

Built-in Provider Catalog

modelIndex selects from the built-in catalog below.

Anthropic

• 0 -> claude-3-haiku-20240307
• 1 -> claude-3-5-haiku-20241022
• 2 -> claude-haiku-4-5
• 3 -> claude-sonnet-4-6
• 4 -> claude-opus-4-7

OpenAI

• 0 -> gpt-5-nano
• 1 -> gpt-5.4-nano
• 2 -> gpt-5-mini
• 3 -> gpt-5.4-mini
• 4 -> o4-mini
• 5 -> gpt-4.1
• 6 -> o3
• 7 -> gpt-5.4
• 8 -> gpt-5.5
• 9 -> o3-pro
• 10 -> gpt-5.5-pro

Gemini

• 0 -> gemini-2.0-flash-lite
• 1 -> gemini-2.5-flash-lite
• 2 -> gemini-2.5-flash
• 3 -> gemini-3-flash-preview
• 4 -> gemini-2.5-pro
• 5 -> gemini-3-pro-preview

DeepSeek

• 0 -> deepseek-chat
• 1 -> deepseek-reasoner
• 2 -> deepseek-v4-pro

Grok

• 0 -> grok-4-1-fast-non-reasoning
• 1 -> grok-4-1-fast-reasoning
• 2 -> grok-code-fast-1
• 3 -> grok-4.20-non-reasoning
• 4 -> grok-4.20-reasoning
• 5 -> grok-4.20-multi-agent

Cost Tracking

When you call prompt(..., true), Rueter AI calculates usage cost from the provider response and returns a UsageCost object:

{
  model: string
  input: string
  output: string
  total: string
}

Values are returned as fixed-precision USD strings.

Specialized Model Factories

The package currently exports 53 specialized model factories. Each one accepts apiKey: string and returns a configured RueterModel.

Examples include:

• CompressorModel
• SimpleAnswerModel
• TerminalCommandModel
• PromptEnhancerModel
• KeywordExtractorModel
• QuestionExtractorModel
• TextFormatterModel
• CodeGeneratorModel
• RefactorModel
• DebugModel
• TestGeneratorModel
• SecurityAuditorModel
• CodeReviewModel
• DocumentationModel
• CommitMessageModel
• CodeExplainerModel
• TypeScriptTyperModel
• PerformanceAnalyzerModel
• FunctionGeneratorModel
• ProjectArchitectModel
• FilePlannerModel
• FileAssemblerModel
• ApiExtractorModel
• DecomposerModel
• PlannerModel
• SelfCritiqueModel
• SummarizerModel
• TranslatorModel
• ConceptExplainerModel
• RootCauseAnalyzerModel
• ArgumentBuilderModel
• DecisionAnalyzerModel
• DataExtractorModel
• JsonResponseModel
• SchemaGeneratorModel
• SentimentAnalyzerModel
• ClassifierModel
• EntityExtractorModel
• DataTransformerModel
• DataValidatorModel
• RegexGeneratorModel
• SQLGeneratorModel
• ApiDesignerModel
• GraphQLSchemaModel
• OpenApiSpecModel
• WritingStyleAnalyzerModel
• StyleReplicatorModel
• AcademicWriterModel
• ResearchOutlinerModel
• EmailDrafterModel
• TechnicalBlogModel
• ChangelogGeneratorModel
• ReadmeGeneratorModel

Example:

import { PromptEnhancerModel } from "rueter-ai"

const model = PromptEnhancerModel(process.env.GROK_API!)
const res = await model.prompt("Make this prompt stronger", true)
console.log(res)

Important: in the current source, all specialized model factories are Grok-backed presets, so they require an xAI/Grok API key.

Workflows

The package exports these workflow functions:

• CodeProjectGenerator
• BugHunterWorkflow
• CodeRefactorWorkflow
• ResearchAssistantWorkflow
• PersonalAuthor

These workflows accept config objects, use the specialized models internally, and usually write files to disk while optionally reporting progress through an onProgress callback.

CodeProjectGenerator

Creates a project from a high-level prompt and writes generated files plus a project hub markdown file.

import { CodeProjectGenerator } from "rueter-ai"

await CodeProjectGenerator({
  apiKey: process.env.GROK_API!,
  projectPrompt: "Build a Node.js CLI expense tracker",
  outputDir: "./ledger",
  projectMdPath: "./LedgerProject.md",
  onProgress(message) {
    console.log(message)
  },
})

BugHunterWorkflow

Scans a source directory and writes a RueterBugReport.md report.

Required config:

• apiKey
• sourceDir

Optional config:

• reportPath
• extensions
• onProgress

CodeRefactorWorkflow

Refactors every source file in a directory, writes the output to another directory, and can also generate tests.

Required config:

• apiKey
• sourceDir
• outputDir

Optional config:

• generateTests
• reportPath
• extensions
• onProgress

ResearchAssistantWorkflow

Generates a research document and writes RueterResearch.md by default.

Required config:

• apiKey
• topic

Optional config:

• outputPath
• sections
• onProgress

PersonalAuthor

Reads or creates RueterAuthor.md, analyzes the writing samples in it, and writes a generated piece to RueterOutput.md by default.

Required config:

• apiKey
• assignment

Optional config:

• authorMdPath
• outputPath
• onProgress

Important: the current workflow implementations use the specialized model presets internally, so they also depend on a Grok/xAI API key in the current source.

GhostWriter

The package also exports:

GhostWriter(
  apiKey: string,
  history: string[],
  prompt: string,
  options?: { log?: boolean }
): Promise<string>

This helper analyzes previous writing samples and then generates a new piece in a similar voice using WritingStyleAnalyzerModel and StyleReplicatorModel.

Example:

import { GhostWriter } from "rueter-ai"

const history = [
  "Sample one",
  "Sample two",
  "Sample three",
]

const output = await GhostWriter(
  process.env.GROK_API!,
  history,
  "Write a new paragraph in this style about solar energy.",
  { log: false }
)

console.log(output)

Pass { log: true } if you want GhostWriter to print the generated style guide and replicated output while it runs.

Exported Types

The package re-exports its public types, including:

• Provider
• ModelInfo
• HttpRequestFormat
• BuilderConfig
• UsageCost
• ModelResult
• RueterModelConfig
• RueterResults

Current Notes and Limitations

• The correct Rueter constructor is new Rueter(models, config). It is not new Rueter({ models: [...] }).
• Rueter requires a config object as its second argument in the current implementation. Pass {} if you do not want shared overrides.
• Rueter.addModel() only accepts provider, apiKey, and modelIndex. If you need per-model config, instantiate RueterModel objects manually and pass them into Rueter.
• Model selection is currently numeric-index based.
• There is no streaming API yet. Responses are returned only after the full provider response is available.
• The core request layer uses native fetch, but the package itself is not literally zero-dependency.

License

MIT