Ultrahost/ docs
All primitives

Intelligence / model routing

LLM Plane

Give your application access to AI. An LLM Plane gives your app a consistent way to use different models, so you can balance quality, speed, and cost without rebuilding the integration every time.

The idea

Imagine building an application that helps customers write better emails. Different AI models have different strengths, speeds, and costs. An LLM Plane sits between your application and model providers, so you can try another model without rebuilding how your app connects to AI.

Interactive demo

AI Writing Studio

Try the same capability with different model profiles.

Model routing preview
Choose a model

Pick a profile, generate, then switch profiles to compare.

Sample response

Model B / Balanced

Generate a response with Model B to see its sample output. Switch models and generate again to compare.

Usage estimate

Approx. tokens—
Response time—
Illustrative cost—

Estimates vary by prompt, output length, and provider. Figures are examples, not live usage.

This is a local simulation with sample responses. It doesn't send your prompt to a model provider, and resets when you refresh.

Behind the scenes

  1. Your application
  2. LLM Plane
  3. AI model provider
  4. Response

Your app sends a request to one stable interface. The plane routes it to a model and returns the response with usage details.

What could you build?

  • Writing assistants
  • AI customer support
  • Document analysis
  • Intelligent business workflows

Good for

  • Multi-model chat and generation
  • Structured extraction
  • Usage, cost, and quality controls

Implementation notes

  • Call a named capability instead of a vendor model.
  • Version prompts alongside your application.
  • Add budgets and fallbacks before production traffic.

Implementing it

With a coding agent

Ask your agent to add AI features

Agent setup (mock link)

Copy this prompt, fill in the bracketed details, and paste it into your coding agent. It describes the goal, the AI capability, and the guardrails.

Bracketed text marks details only you can fill in. 783 / 6,000 characters.

Fill in:APP OR PRODUCTCAPABILITY NAMEINPUTOUTPUT FORMATBUDGET OR USAGE LIMITWHEN A MODEL IS SLOW OR UNAVAILABLEUSER, WORKSPACE, OR FEATUREIMPORTANT PROMPTS, BAD OUTPUTS, AND LIMIT CASES

Good foundation · 68/100

Traditional SDK / API

Use it from a function

TypeScript
const result = await ultrahost.llm.generate({
  capability: "support-reply",
  input: ticket,
  responseFormat: "json",
})

Key takeaway

The LLM Plane gives your application a consistent way to work with AI models while managing usage and costs.

Next primitive

Vector DB